# Grok Bot — full corpus # LLM Wiki An open-source template for building LLM-powered knowledge bases, following [Andrej Karpathy's "LLM Wiki" pattern](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f). You provide raw sources. The LLM reads them, writes structured wiki pages, cross-links everything, and maintains it over time. You never edit the wiki directly — you curate sources and ask questions. ## How It Works The system has three layers: ``` raw/ Sources you collect (articles, transcripts, notes, PDFs) wiki/ LLM-written & maintained pages (summaries, concepts, entities, syntheses) CLAUDE.md Schema that tells the LLM how to structure everything ``` Three operations drive the workflow: | Operation | Trigger | What happens | |-----------|---------|--------------| | **Ingest** | "ingest raw/my-source.txt" | LLM reads the source, creates a summary page, creates/updates concept and entity pages, adds cross-links, updates the index and log | | **Query** | Ask any question | LLM searches the wiki, synthesizes an answer with citations, optionally creates a synthesis page for novel insights | | **Lint** | "lint" or "health check" | LLM audits all pages for orphans, contradictions, missing links, incomplete sections, and low-confidence claims — fixes what it can, reports the rest | ## Quick Start 1. **Clone this repo** ```bash git clone https://github.com/YOUR_USERNAME/llm-wiki.git my-knowledge-base cd my-knowledge-base ``` 2. **Customize CLAUDE.md** for your domain - Update the Purpose section with your topic - Replace the placeholder tagging taxonomy with your own categories - Adjust confidence level descriptions if needed - Everything else (workflows, page formats, linking rules) works as-is 3. **Drop sources into `raw/`** - Text files, transcripts, articles, notes — any plain text - These are immutable once added; the LLM never modifies them 4. **Tell the LLM to ingest** ``` ingest raw/my-first-source.txt ``` The LLM will create summary pages, concept pages, entity pages, cross-links, and update the index. 5. **Ask questions** ``` What are the key differences between X and Y? ``` The LLM answers from the wiki, citing specific pages. 6. **Run health checks** ``` lint ``` The LLM audits the wiki and fixes issues. ## Directory Structure ``` . ├── CLAUDE.md # Schema — the LLM's instructions ├── raw/ # Your source documents (immutable) └── wiki/ ├── index.md # Master catalog of all pages ├── log.md # Append-only activity log ├── dashboard.md # Dataview dashboard (Obsidian) ├── analytics.md # Charts View analytics (Obsidian) ├── flashcards.md # Spaced repetition cards ├── summaries/ # One page per source document ├── concepts/ # Concept and framework pages ├── entities/ # People, tools, organizations, etc. ├── syntheses/ # Cross-cutting analyses and comparisons ├── journal/ # Research/session journal entries │ └── template.md # Journal entry template └── presentations/ # Marp slide decks ``` ## Enhancements This template includes several extras beyond the core wiki pattern: ### Dataview Dashboard (`wiki/dashboard.md`) Live queries that surface low-confidence pages, recent updates, concepts by tag, and pages with the most sources. Requires the [Dataview](https://github.com/blacksmithgu/obsidian-dataview) Obsidian plugin. ### Charts View Analytics (`wiki/analytics.md`) Visual analytics with pie charts, bar charts, and word clouds. Requires the [Charts View](https://github.com/caronchen/obsidian-chartsview-plugin) Obsidian plugin. ### Mermaid Diagrams Use Mermaid code blocks in any wiki page to create flowcharts, sequence diagrams, or concept maps. Native support in Obsidian and GitHub. ### Marp Slides (`wiki/presentations/`) Create slide decks from markdown using [Marp](https://marp.app/). Drop presentation files in this directory. ### Research Journal (`wiki/journal/`) Track your research sessions, experiments, or applied work with the included template. The LLM can reference journal entries when answering queries. ### Spaced Repetition (`wiki/flashcards.md`) Flashcards in the format used by the [Spaced Repetition](https://github.com/st3v3nmw/obsidian-spaced-repetition) Obsidian plugin. Ask the LLM to generate flashcards from any wiki page. ### MCP Server This repo works with Claude Code's MCP server capabilities. Point an MCP-compatible client at this repo and the LLM can read/write the wiki programmatically. ## Customizing for Your Domain The schema in `CLAUDE.md` is domain-agnostic. To adapt it: 1. **Purpose** — Describe your knowledge domain in one paragraph 2. **Tagging taxonomy** — Replace placeholder categories with your own (e.g., for a cooking KB: `cuisine`, `technique`, `ingredient`, `equipment`) 3. **Confidence levels** — Adjust the descriptions to match your domain's evidence standards 4. **Entity types** — Update the entity page description to match what entities mean in your domain (people, tools, companies, etc.) 5. **Journal template** — Customize `wiki/journal/template.md` for your workflow Everything else — page format, linking conventions, workflows, rules — is universal and works across domains. ## Example Domains This template works for any knowledge-intensive topic: - **Research notes** — papers, experiments, methodologies - **Book analysis** — themes, characters, author techniques - **Competitive analysis** — companies, products, market trends - **Course notes** — lectures, readings, key concepts - **Personal development** — frameworks, habits, book summaries - **Technical documentation** — APIs, architectures, design patterns - **Hobby deep-dives** — any subject you want to master ## License MIT --- title: "Grok Bot KB — Master Index" type: index updated: 2026-10-21 --- # Grok Bot KB — Master Index **Domain:** Grok Bot, the persistent cloud-computer AI teammates from xAI (styled "SpaceXSI" on its own pages; "SpaceXAI" in press) and Cursor, in beta since 2026-08-11: bots and team bots, messaging and @bot on X, the bot's computer, connectors, skills and routines, approvals, privacy, settings, and enterprise identity, networking, proxies, deployment and inference hooks. **Scope:** Built from both official doc sites (docs.x.ai/grok-bot and cursor.com/docs/grok-bot) fetched 2026-10-21, the x.ai plans news post and the xAI release notes, plus three attributed third-party write-ups. Where the two doc sites disagree, pages show both. Not covered: plan prices and allowances (the plans help page was not fetched), egress IP ranges, admin API endpoint details, SCIM/SAML field mappings. **Pages:** 28 (5 syntheses · 20 concepts · 1 entities · 2 summaries) **Sources fetched:** 2026-10-21 Start at [[syntheses/overview]] if you are new. ## Syntheses - [[syntheses/enterprise-rollout-guide]] — Ordered admin checklist from plan to pilot (identity, network/proxy, deployment, policies), linking each step to its page. - [[syntheses/getting-started]] — This page merges the two official get-started guides, docs.x.ai/grok-bot/get-started.md and cursor.com/docs/grok-bot/get-started.md, into one first-run sequence. - [[syntheses/overview]] — Grok Bot is a product for running Bots: named AI teammates with a job, a memory and their own persistent cloud computer (browser, filesystem and terminal). - [[syntheses/security-model]] — Threat model and controls table across both hosts' security, approvals, identity, network and proxy docs; which defaults are permissive and what admins must turn on. - [[syntheses/troubleshooting]] — Symptom → cause → fix table from the troubleshooting and FAQ pages, mapped to the relevant wiki pages. ## Concepts - [[concepts/approvals]] — Approvals keep consequential Bot actions under the member's control. - [[concepts/apps-and-connectors]] — How a Bot reaches outside services: signing into apps in its browser, and connectors/MCP; the Apply to MCP servers switch. - [[concepts/bots]] — A Bot is a durable AI teammate with a name, a job, its own conversation, and working context that builds up over time. - [[concepts/cloud-computer]] — Every Grok Bot user gets one persistent, Cursor-hosted cloud computer, and every Bot on that account does its work there: browser, command line, files, and connected… - [[concepts/enterprise-deployment]] — How an IT team installs the Grok Bot client apps on member devices with its existing device-management tools, pins versions, and plans updates. - [[concepts/files-and-results]] — This page covers how you give a Bot source material (attachments, links, shared files) and how its outputs come back to you: cards in the conversation, files in the… - [[concepts/identity-and-access]] — How people get into Grok Bot and how Bots act under their identity. - [[concepts/inference-hooks]] — Inference hooks send every governed Grok Bot model request, signed, to an AI security server that your organization runs or uses through a security vendor. - [[concepts/messaging-and-collaboration]] — Working with Grok Bot is meant to feel like messaging a teammate. - [[concepts/mobile-app]] — The Grok Bot mobile app is a client for iPhone, iPad and Android. - [[concepts/plans-and-access]] — Grok Bot is not sold as a standalone plan in the official docs. - [[concepts/privacy-and-data]] — Grok Bot uses Cursor authentication and Cursor account data settings. - [[concepts/private-networks]] — How Bots reach internal systems (internal APIs, source control, databases, staging) from hosted computers that run in Cursor's cloud and egress through shared static… - [[concepts/proxies]] — Gateway configuration for the path from a member's device to Cursor. - [[concepts/settings-and-notifications]] — The Grok Bot settings dialog holds app-wide and desktop-local behavior. - [[concepts/skills-routines-and-automations]] — Skills (reusable instructions, Teach a task) and routines (scheduled or triggered runs), with limits and UI names per host. - [[concepts/tag-bot-on-x]] — When you tag @bot (https://x.com/bot) in a post or reply on X, that post is handed to your Grok Bot as a task. - [[concepts/team-bots]] — A Team Bot is one Bot that the whole team talks to. - [[concepts/teams-and-enterprises]] — How Grok Bot is made available to a Cursor team and which controls admins get. - [[concepts/use-cases]] — The official use-case pages (docs.x.ai use-cases.md and cursor.com use-cases.md) describe eight Bot roles. ## Entities - [[entities/grok-bot]] — Grok Bot is a product for creating Bots: persistent, named AI teammates. ## Summaries - [[summaries/docs-map]] — Catalog of all 42 raw sources by host, with topics documented on only one host and the divergences between the xAI and Cursor docs. - [[summaries/release-history]] — Dated timeline from the release notes, the plans news post and attributed coverage; there is no dedicated Grok Bot changelog. --- title: "Approvals and Auto Review" type: concept tags: [approvals, auto-review, ask-first, allow-automatically, local-execution, security] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-approvals-security-and-privacy-md.md", "raw/web_community-security-md.md", "raw/web_community-security-md-2.md", "raw/web_community-security-faq-md.md", "raw/web_community-security-faq-md-2.md", "raw/web_community-teams-md.md", "raw/web_community-teams-and-enterprises-md.md", "raw/web_community-settings-md.md", "raw/web_community-troubleshooting-md.md", "raw/web_community-team-bots-md.md", "raw/web_community-faq-md.md", "raw/web_community-work-md.md"] --- # Approvals and Auto Review ## Definition Approvals keep consequential Bot actions under the member's control. There are three layers: 1. **Boundaries in the request.** The docs call this "the strongest boundary". Example: "Do not change the campaign or message the agency. Ask for approval after showing the current value, proposed value, and expected impact." 2. **Approval cards** for proposed actions. 3. **Auto Review**, an independent review model that evaluates risky actions *before they run* and can let an action proceed, require approval, or deny it. **Local-computer execution** has its own separate policy (below). ## How It Works **Approval card.** The card shows the proposed operation and its inputs. - **Allow once** lets the Bot continue with that action. - **Always allow** can save a matching rule. - **Deny** blocks the action. The controls are the same on iPhone. An approval controls the proposed action. It **does not reverse work already completed**. **Timeouts.** An Auto Review approval from a chat you are in **waits for your answer**. An approval from work started without you (a routine, a trigger, or a message from another Bot) **expires after about 10 minutes**. The card then shows **Expired** and the action does **not** run. An expired card may offer **Always allow this in the future**, which adds the rule without running the expired action. To finish the task, ask the Bot to try again (docs.x.ai). **What Auto Review covers** (xAI `security.md` and both security FAQs; docs.x.ai's approvals page says only that Auto Review "evaluates tool calls and computer actions"): shell commands, app (plugin) calls, computer use, automation writes (changes to routines and event triggers), and delegation such as Cloud Agent and subagent launches. It does **not** review every side effect. Memory writes and most settings changes are examples it misses. **Rules.** Members set them in **Settings → General → Auto-review** (cursor.com: in **Custom Rules**; xAI troubleshooting: **Settings → General → Bot → Auto-review**). - **Ask first** rules always stop matching actions. - **Allow automatically** rules let matching actions proceed *only when* the reviewer finds no other reason to stop. - If both kinds match, **Ask first wins**. **Team enforcement (Enterprise only).** - **Enforce Auto-review** is **off by default**. When it is on, members can't turn Auto-review off, unless a group they belong to uses **Don't enforce for this group**. - Team **Auto-review rules** apply to every member's Bots and appear as locked rows (`Required by your admin. You can't edit or delete this rule.`). Members' personal rules can only make behavior *stricter*. - If enforcement is turned off, team rules stop applying. **Team Bots.** In a conversation where nobody can answer an approval card, the Bot runs **without Auto-review** and stays within its set-up permissions, unless your team requires Auto-review (`team-bots.md`). See [[concepts/team-bots]]. ## Key Parameters | Setting | Default | Notes | |---|---|---| | Unattended approval expiry | ~10 minutes | Only for work started without you | | Enforce Auto-review (team) | **Off** | Enterprise only | | Member **Execution on Local Computer** | **Ask every time** (per-command approval) | Options: **Ask every time**, **Always allow**, **Never allow** | | Team **Execution on Local Computer** ceiling | **Always allow** (leaves the choice to each member) | **Ask every time** forces approval for every task. **Never allow** turns local execution off team-wide. A stricter member setting still applies | | **Use hardware security keys** | **On** on macOS and Windows. Not yet supported on Linux | Every use asks for approval (docs.x.ai) | **Local execution path.** In xAI's docs it is **Settings → General → Bot → Execution on Local Computer**. It moves to **Settings → Computer → Computers** (per-computer **Execution on this computer**) once you have registered computers. Cursor's docs place it under **Settings → Computer → Your computer**. The first request shows **Allow Grok Bot and all Bots to run commands on your local computer?** with **Always allow**, **Allow once**, **Never**, and **Deny once** (Esc). **Always allow** and **Never** set the policy for *every* Bot. **Always allow** is unavailable if an admin has set a stricter ceiling. **Source conflict (flagged).** The two hosts disagree on where personal Auto-review rules are stored: - docs.x.ai (`approvals-security-and-privacy.md`, `security.md`): stored **on the current desktop** and synced to its computer, so "another desktop installation needs its own". - cursor.com (`security.md`, `settings.md`): saved **to the member's account**, so they apply "from every desktop and the mobile app". ## When To Use - Write narrow rules tied to a known action. For example: "Ask first before sending any external email", or "Allow automatically when running `git status` in `/workspace/reports`". Avoid "allow everything in the browser". - Keep sending, publishing, purchasing, deleting, permission changes, production changes, and legal-terms acceptance behind approval. - Choose **Never allow** for local execution unless a Bot needs local files. ## Risks & Pitfalls - Auto Review is **model-based**. Use it alongside least privilege, the network policy, and per-user isolation, which don't depend on a model's judgment. It does not replace them. - Never approve an action whose target or effect you can't identify. Ask the Bot for a plain-language explanation or a draft first. - Reactions shouldn't carry a safety-critical decision. Reply in the thread instead. - "Stop now" halts work but doesn't undo actions already completed. - **Action keeps needing approval?** Look for a matching **Ask first** rule, including admin-required team rules. If a card is no longer actionable: reject or cancel it, send a replacement instruction, and ask the Bot to regenerate the action with the corrected scope (see [[syntheses/troubleshooting]]). - Local-execution settings don't stop the Bot from using its **cloud** computer. ## Related Concepts - [[syntheses/security-model]], [[concepts/privacy-and-data]], [[concepts/inference-hooks]] - [[concepts/teams-and-enterprises]], [[concepts/settings-and-notifications]] - [[concepts/cloud-computer]], [[concepts/apps-and-connectors]] ## Sources - raw/web_community-approvals-security-and-privacy-md.md (https://docs.x.ai/grok-bot/approvals-security-and-privacy.md) - raw/web_community-security-md.md (https://cursor.com/docs/grok-bot/security.md) - raw/web_community-security-md-2.md (https://docs.x.ai/grok-bot/security.md) - raw/web_community-security-faq-md.md (https://cursor.com/docs/grok-bot/security-faq.md) - raw/web_community-security-faq-md-2.md (https://docs.x.ai/grok-bot/security-faq.md) - raw/web_community-teams-md.md (https://cursor.com/docs/grok-bot/teams.md) - raw/web_community-teams-and-enterprises-md.md (https://docs.x.ai/grok-bot/teams-and-enterprises.md) - raw/web_community-settings-md.md (https://cursor.com/docs/grok-bot/settings.md) - raw/web_community-troubleshooting-md.md (https://docs.x.ai/grok-bot/troubleshooting.md) - raw/web_community-team-bots-md.md (https://docs.x.ai/grok-bot/team-bots.md) - raw/web_community-faq-md.md (https://docs.x.ai/grok-bot/faq.md) - raw/web_community-work-md.md (https://cursor.com/docs/grok-bot/work.md) --- title: "Apps, Connectors, MCP, and Browser Use" type: concept tags: [connectors, plugins, mcp, oauth, browser, sign-in, connect-apps] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-computer-and-apps-md.md", "raw/web_community-work-md.md", "raw/web_community-security-md.md", "raw/web_community-security-faq-md.md", "raw/web_community-security-faq-md-2.md", "raw/web_community-teams-md.md", "raw/web_community-team-bots-md.md", "raw/web_community-settings-md.md", "raw/web_community-identity-md.md", "raw/web_community-troubleshooting-md.md", "raw/web_community-official-grok-bot-breakdown-2026.md"] --- # Apps, Connectors, MCP, and Browser Use ## Definition A Bot reaches outside services in two ways: 1. **Connectors**, which give it a structured, tool-based integration. The docs.x.ai pages call them "plugins". The cursor.com pages call them "apps" or "connected apps" (MCP servers). 2. **The browser** on its [[concepts/cloud-computer]], which it drives like a person. This covers services that have no connector. Both hosts say to prefer a connector when one exists. docs.x.ai says it is "often more reliable than clicking through a website"; cursor.com says apps are "more reliable than clicking through a website when one exists". ## How It Works ### Connecting an app (docs.x.ai steps) 1. Open **Connect apps** from the sidebar. You can also follow an in-chat **Connect** card (cursor.com). 2. Browse the available plugins. 3. Choose **Add**. 4. Complete authentication in your browser if requested. 5. In chat, type `@` to attach the connector to a task. Type `/` to reference a saved skill. The connected count next to the **Connect apps** title manages installed apps and **Private skills**. You can enable or disable individual app tools there (cursor `settings.md`). Apps are not in Settings. ### Where tokens live - OAuth tokens are held on **Cursor's connector backend**. The Bot invokes tools without receiving them, and tokens are never stored on the computer. - Connecting an app does **not** go through the computer browser, so IdP device-trust rules for the computer browser don't apply to connector sign-in (see [[concepts/identity-and-access]]). ### Scope - **Installed connectors are account-wide.** Every Bot you run can use them. They are not isolated to one Bot. - **Team Bots** use OAuth plugins with the access of *the person talking to the Bot*. Key or token plugins use the Bot's own credential. In a shared conversation (a Slack thread or group chat), a Team Bot **asks first** before it uses your personal connector. The options are **Allow once**, **Always allow for this Bot**, **Always allow for all Team Bots**, and **Skip**. To undo remembered answers, use **Settings → General → Team Bots → Clear**. See [[concepts/team-bots]]. - Team Bot custom MCP servers come in two kinds. **Remote HTTPS** uses the Bot's credential or each person's OAuth. **Command** runs on whichever computer the conversation uses, so never put a credential in its command or arguments. ### Browser use and sign-in - Browser sessions persist, so one sign-in usually lasts across tasks, and it is shared with all your Bots. - For a password, passkey, 2FA, CAPTCHA, payment or identity check, or a site that requires a human, the Bot asks you to **take over**: choose **Take over** or **Take control**, complete only the blocked step, choose **I'm done**, then tell the Bot to continue. The Bot doesn't type credentials. Never paste passwords or one-time codes into chat. - For supported connections, a **secure secret request** masks the value, keeps it out of the transcript, and keeps it from the model. ### X without connecting Bots can search and read **public** X content without an X connection (docs.x.ai). That access is read-only and doesn't sign in as you. For DMs, bookmarks, the home timeline, or likes, add the X plugin from **Connect apps**. ## Key Parameters | Item | Value / behaviour | |---|---| | Blocked by team policy | The app shows **Disabled by team admin** | | Where admins set the connector policy | Dashboard **Plugins & MCPs** page (Team Marketplace), not the Grok Bot page | | Separate Grok Bot connector list | None. Grok Bot inherits the Cursor connector policy | | MCP allowlist | Enterprise only | | Admin push or default-on connectors | **Not available** | | **Apply to MCP servers** (cursor.com security) | Off by default. Appears only in an allowlist network mode. When on, remote MCP servers, including built-ins such as GitHub (`api.github.com`), X (`api.x.com`), and Origin (`api.origin.cursor.com`), must match **Allowed destinations** by host | ## When To Use - **Connector:** use one for Gmail, Notion, Slack, and other supported services when it exposes what you need. - **Browser:** use it for services without a connector, or for visual workflows a connector doesn't expose. - **Take over:** use it for any credential or human-verification step. Ask the Bot to pause and notify you rather than bypass a re-verification. ## Risks & Pitfalls - **Blocking an app doesn't block the service's website.** Closing the browser path needs **Network Controls**, which is Enterprise only. See [[syntheses/security-model]]. - An allowlist network mode by itself does **not** restrict remote MCP servers. Bots reach them from Cursor's servers. You need **Apply to MCP servers** (cursor.com). A connector whose hosts aren't all allowed is refused with `MCP server "" refused: it is not on your team's Grok Bot network allowlist (Grok Bot settings > Network > Allowed destinations). A team admin can add it.` - **Plugin won't install or authenticate:** check **Connect apps → Your plugins → Installed**, re-run the authentication, and check for org variables or admin configuration. If the source service revoked access, remove and reconnect. If a permitted app fails with a vendor permission error, some vendors restrict their MCP endpoints to their own admins. - Some sites expire sessions or re-verify every sensitive action. That can't always be avoided. - **Third-party claims (Vellum, a competing-product vendor; medium confidence):** custom MCP servers "must be publicly reachable", so a local server needs tunneling. (The official Team Bots doc separately describes **Command** MCP servers that run on the computer.) Vellum also lists built-in connectors as Gmail, Google Calendar, Google Drive, OneDrive, Outlook Mail and Calendar, Microsoft Teams, SharePoint, and Salesforce, and calls them "fixed". The official host table in cursor.com `security.md` names more, including Google Docs/Sheets/Slides/Ads, Stripe Link, Tailscale, QuickBooks Online, Workday, WhatsApp Business, and Zoho. Treat Vellum's list as incomplete. ## Related Concepts - [[concepts/cloud-computer]], [[concepts/approvals]], [[concepts/identity-and-access]] - [[concepts/team-bots]], [[concepts/skills-routines-and-automations]] - [[syntheses/security-model]], [[syntheses/troubleshooting]] ## Sources - raw/web_community-computer-and-apps-md.md (https://docs.x.ai/grok-bot/computer-and-apps.md) - raw/web_community-work-md.md (https://cursor.com/docs/grok-bot/work.md) - raw/web_community-security-md.md (https://cursor.com/docs/grok-bot/security.md) - raw/web_community-security-faq-md.md (https://cursor.com/docs/grok-bot/security-faq.md) - raw/web_community-security-faq-md-2.md (https://docs.x.ai/grok-bot/security-faq.md) - raw/web_community-teams-md.md (https://cursor.com/docs/grok-bot/teams.md) - raw/web_community-team-bots-md.md (https://docs.x.ai/grok-bot/team-bots.md) - raw/web_community-settings-md.md (https://cursor.com/docs/grok-bot/settings.md) - raw/web_community-identity-md.md (https://cursor.com/docs/grok-bot/identity.md) - raw/web_community-troubleshooting-md.md (https://docs.x.ai/grok-bot/troubleshooting.md) - raw/web_community-official-grok-bot-breakdown-2026.md (https://www.vellum.ai/blog/official-grok-bot-breakdown), third-party --- title: "Bots: Create, Configure, and Manage" type: concept tags: [grok-bot, bots, memory, sidebar, templates, lifecycle] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-bots-md.md", "raw/web_community-work-md.md", "raw/web_community-settings-md.md", "raw/web_community-teams-md.md", "raw/web_community-settings-and-notifications-md.md"] --- # Bots: Create, Configure, and Manage ## Definition A **Bot** is a durable AI teammate with a name, a job, its own conversation, and working context that builds up over time. Both official hosts (docs.x.ai `bots.md` and cursor.com `work.md`) use the same definition. Each Bot has its own role and conversation, but **all Bots on one account share one cloud computer** (see [[concepts/cloud-computer]]). They can pass context to each other through direct messages, group chats, and shared files. ## How It Works ### Give each Bot one clear job Create a separate Bot when the work has its own goal or area of ownership, set of tools and sources, working style, approval boundary, or recurring schedule. The docs give **Talent Scout**, **Expense Manager**, and **Bug Reproduction** as good jobs. A **General Helper** gives the Bot less guidance and makes its saved context harder to reuse. Describe the role in operational terms: what it owns, what it produces, and what it must never do without approval. ### Create a Bot The two hosts describe the flow differently: | Step | docs.x.ai (`bots.md`) | cursor.com (`work.md`) | |---|---|---| | Start | **New** in sidebar or `Cmd/Ctrl+N`, then in **New chat** choose **Create new Bot**, or type a name and choose **Create “name” Bot** | **New chat** (the **+** button) or Cmd+N, then **Create new Bot** | | Result | Opens a Bot named **New Bot** | (not stated) | | Identity | Bot menu or **Edit Profile** sets name, label, description, avatar | Click the Bot's name at the top of the chat to rename, add a label, or change the picture | | First message | Start with a concrete task | Give the job, sources, and approval boundaries, ask it to remember them, and send a real first task with a clear finish line | Existing Bots can also suggest or create a focused Bot when a job needs a long-lived owner. If you want a small roster, ask them to check with you before creating several. ### Lasting rules versus task instructions - docs.x.ai: put lasting rules in the **Description** (for example, “Never send external messages without approval.”) and put task-specific instructions in messages. - cursor.com: **personal Bots don't have a description** (`settings.md`). To change what a Bot does, tell it in chat and ask it to remember (“From now on, never send external messages without approval.”). Only a [[concepts/team-bots]] Team Bot has a description, which its owner or a Manager edits under **About**. **Divergence:** the two hosts disagree on whether personal Bots have a description field. As of fetch 2026-10-21, cursor.com is the more specific source. ### Memory A Bot keeps stable working preferences, important facts, and summaries of its work, so it can stay in its role without replaying every earlier message. Memory does **not** replace an authoritative source. Keep changing facts in the source system, ask the Bot to cite or reopen current data before consequential decisions, and correct stale assumptions directly. cursor.com adds that safety boundaries should be backed by approvals and Auto-review rules, not by memory alone (see [[concepts/approvals]]). ### Organize, duplicate, share, delete - **Pin** keeps active Bots at the top of the sidebar. - **Sections** (cursor.com only): right-click a Bot, then **Move to new section**, or **Move to** once sections exist. Sections sync between desktop and iOS. - **Hide from sidebar** removes a Bot from the main list without deleting its work. To restore it, open **Hidden Bots** and choose **Show in sidebar**. On iPhone and Android, choose **Unhide**. When every Bot is hidden, the mobile home screen shows **Show Hidden Bots**. **Hiding does not pause the Bot or its routines.** - **Duplicate** makes a copy named "`` copy" that carries the profile, settings, enabled skills, routines, and avatar. It does **not** copy conversation history, learned memory, or chat attachments. cursor.com describes Duplicate as an iPhone action. - **Share** a template from the **Share menu** with **Create template**. When the template is ready, **Copy link**, **View template details**, and **Update template** appear. On docs.x.ai you then choose **Public link** or **Team-only**: Enterprise accounts default to **Team-only** and other accounts default to a public link. The recipient previews the template on x.ai and chooses **Add to Grok Bot**, which creates a copy on their account. They never get your computer, logins, or history. Adding a shared Bot accepts the third-party bot terms. - **Marketplace** in the sidebar lists Bots you can add with **Add** or **Import Bot**. Either option creates your own copy from the Bot's template. - **Delete** removes the Bot's profile, conversation, and routines. Shared computer files and sign-ins are **not** isolated per Bot and may stay on the computer, so clean them up separately. docs.x.ai notes that backend retention follows the applicable Cursor terms. ## Key Parameters | Item | Value / name | |---|---| | New Bot shortcut | `Cmd/Ctrl+N` | | Default new-Bot name (docs.x.ai) | **New Bot** | | Profile fields | name, label, description (docs.x.ai; Team Bots only per cursor.com), avatar | | Duplicate name | "`` copy" | | Template link scopes | **Public link**, **Team-only** (Enterprise default **Team-only**) | | Admin control | **Public template sharing**. Enterprise teams start with it off; other teams start with it allowed. Cursor enforces it server-side, including on templates that are already public (`teams.md`). | | Per-Bot settings path | **View conversation details** → **Bot settings** (see [[concepts/settings-and-notifications]]) | ## When To Use - Use a **new Bot** for a distinct, recurring role with its own tools, boundary, or schedule. - Use **Duplicate** for the same role with a different scope, for example one account-health Bot per region. Rename the copy and give it the new scope before assigning work. - Use a **template** when someone else should start from your Bot's configuration. Use a **Team Bot** when the whole team should talk to one shared Bot. - **Hide** rather than delete when you may need the work later. - Recommended roster discipline: give one Bot ownership of an end-to-end outcome, add specialists only for stable roles, use a group chat when handoffs should be visible, and keep external actions behind approval (see [[concepts/messaging-and-collaboration]]). ## Risks & Pitfalls - **Public templates expose configuration.** A public link shows the Bot's identity, description, skills, and routines to anyone who has it. Remove API keys, internal URLs, and customer data before sharing. - **Deleting is not a full cleanup.** Files and sign-ins on the shared computer can stay after a Bot is deleted. Deleting a Bot also deletes the routines it owns. - **Hidden does not mean paused.** Routines on a hidden Bot keep running. - **Bots are not security boundaries from each other.** Any of your Bots can use a credential or file on the computer. - **Stale memory.** Memory can hold outdated facts. Ask the Bot to reopen sources for consequential decisions. - **Vague roles.** A "General Helper" Bot builds up context that is hard to reuse. ## Related Concepts - [[concepts/team-bots]]: shared Bots with team memory and managers - [[concepts/messaging-and-collaboration]]: group chats and Bot-to-Bot handoffs - [[concepts/skills-routines-and-automations]]: the skills and routines a Bot owns - [[concepts/settings-and-notifications]]: per-Bot settings - [[concepts/cloud-computer]]: the computer all your Bots share - [[concepts/approvals]]: enforcing boundaries - [[entities/grok-bot]] ## Sources - `raw/web_community-bots-md.md`: https://docs.x.ai/grok-bot/bots.md - `raw/web_community-work-md.md`: https://cursor.com/docs/grok-bot/work.md - `raw/web_community-settings-md.md`: https://cursor.com/docs/grok-bot/settings.md - `raw/web_community-teams-md.md`: https://cursor.com/docs/grok-bot/teams.md - `raw/web_community-settings-and-notifications-md.md`: https://docs.x.ai/grok-bot/settings-and-notifications.md --- title: "The Bot's Cloud Computer" type: concept tags: [cloud-computer, firecracker, persistence, workspace, recovery, admin, computers] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-computer-and-apps-md.md", "raw/web_community-work-md.md", "raw/web_community-computers-md.md", "raw/web_community-computers-md-2.md", "raw/web_community-security-md.md", "raw/web_community-security-md-2.md", "raw/web_community-security-faq-md.md", "raw/web_community-teams-md.md", "raw/web_community-private-networks-md.md", "raw/web_community-identity-md.md", "raw/web_community-deployment-md.md", "raw/web_community-settings-md.md", "raw/web_community-troubleshooting-md.md", "raw/web_community-team-bots-md.md", "raw/web_community-official-grok-bot-breakdown-2026.md", "raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md"] --- # The Bot's Cloud Computer ## Definition Every Grok Bot user gets **one persistent, Cursor-hosted cloud computer**, and every Bot on that account does its work there: browser, command line, files, and connected tools. It is assigned to the *user account*, not to an individual Bot. Work keeps running when the desktop app, laptop, or phone is closed. Both official hosts say this: `security.md` on each host says Grok Bot "runs only on Cursor-hosted cloud computers", cursor.com `teams.md` says "One persistent Firecracker microVM per user", and `computer-and-apps.md` / `work.md` say the computer belongs to the account and is shared by every Bot on it. ## How It Works **Isolation.** Each user's computer is a dedicated **Firecracker microVM** with its own kernel, memory, and virtual devices. One user cannot reach another user's computer (security FAQ and `teams.md`, both hosts). **Shared within a user.** All of a user's Bots share browser cookies and sessions, files, and command-line credentials, so one Bot can pick up work another saved. Each Bot gets **its own screen**. Several Bots can use browser and desktop tools in parallel, but one Bot runs only **one computer-use task on its screen at a time**. The docs say screens are "work surfaces, not security boundaries". **Operating system and tooling (official statements only).** - "The computer runs Linux and isn't enrolled in MDM", and device-trust agents such as Okta FastPass don't run on it (`identity.md`). - Team Setup scripts assume a client that runs on Debian-based Linux. Scripts run as the computer user, with `sudo` available (`private-networks.md`). - **Allow Sudo on Bot Computers** (Enterprise) lets Bots run commands as root. It is **on by default**, and existing computers pick up a change only after they restart (`teams.md`). - CPU, RAM, and disk sizes are **not documented** in the corpus. **Persistence.** - The shared workspace is at `/workspace`. Files, browser state, and supported sign-ins are designed to survive normal updates and recovery. Temp directories, manually installed packages, and uncommitted app state are replaceable. - A durable disk keeps local files, browser sessions, and browser-saved data across sessions. Connector OAuth tokens are **never stored on the computer** (they stay on Cursor's connector backend). - Idle computers **hibernate automatically**. Hibernation is not deletion. - Conversations are stored outside the computer, so they survive even a reset (cursor `work.md`). **Team Bots.** The owner's chat runs on the owner's computer. A teammate's app chat or 1:1 Slack DM runs on *that teammate's* computer. Slack channels, group DMs, and threads share one separate computer for that Bot (`team-bots.md`). **Hosting.** The computers run only on Cursor-hosted infrastructure, in the United States today. On-prem, inside-your-perimeter, and bring-your-own-image deployments are not supported. Cursor maintains the image, so you cannot pin or supply one (`deployment.md`). ## Key Parameters | Control | Where | What it does | |---|---|---| | **Update Bot's Computer** | Settings → Computer → Bot's computer | Moves the computer to the latest version. A *software* update keeps installed apps. A *computer* update rebuilds the computer and keeps files and logins, but removes installed apps and packages (cursor `settings.md`). | | **Reset Bot's Computer** | same section | A last resort. Rebuilds from the last saved data or snapshot, and very recent changes can be lost. | | **Recover computer** / **Recover Grok Bot's Computer** | The unreachable-computer error state only, not Settings | Rebuilds from the last saved data. | | **Reset Computer** (mobile) | iPhone/Android: Settings → Bot → Bot Computer | The mobile reset path (xAI `troubleshooting.md`). | | **Manage Member Computers and Data** | Dashboard → Grok Bot → Security & Compliance | Enterprise only. Offers **Recreate VMs**, **Terminate VMs**, **Delete VMs and Data**, and **Delete All Grok Bot Data**. | | **Terminate Inactive Computers** | Dashboard → Bot Computers | **Off by default.** Terminates computers after 30 days without use, and computers already past 30 days are included. | | **Restart all VMs** | Banner **Restart VMs to apply changed settings** | Enterprise team admins can recreate every team computer after a settings change (xAI `computers.md` only). | | Organization API | `admin:*` scope key | Lets a script recreate, terminate, or delete computers. One operation per team at a time, shared with the dashboard. | Recreate and Terminate both **keep the durable disk**, so synced Bots, files, and logins come back, but they remove apps and packages the member installed. Team Setup installs come back. Terminate **does not remove access**. **Delete All Grok Bot Data** is permanent, handles at most 100 members per operation, and is recorded in the Audit Log. On xAI's host, both delete actions require you to type `DELETE` to confirm. Analysis: the two hosts word "Reset" differently. xAI says "keeps the synced durable data" and "last saved snapshot". Cursor says "rebuilds the computer from the last saved data". They agree that recent unsynced work can be lost. ## When To Use - Keep durable project output in `/workspace` or attach it to the conversation. - Use the recovery ladder in order: retry or reopen, restart the app, **Recover**, **Update Bot's Computer**, and **Reset** only last. See [[syntheses/troubleshooting]]. - Admins rolling out a new image or Team Setup manifest use **Recreate**. To end a stuck computer or a member's current work, use **Terminate**. See [[concepts/enterprise-deployment]]. ## Risks & Pitfalls - **Bots are not a security boundary.** Any credential or file on the computer is usable by every Bot on the account. If a workload needs its own credentials, give it its own Cursor user. See [[syntheses/security-model]]. - Deleting a Bot does **not** remove computer files or browser sessions. - A recreate can drop in-computer sign-in sessions, so members sign in again. A turn that was running when a computer was terminated is lost. - A recreate fails for a member if a mid-turn Bot cannot pause in time. The member keeps their current computer. - Inactive-termination has no grace period, and it applies if *any* of the member's Enterprise teams turns it on. - Some services flag datacenter IPs. Egress uses shared static ranges, and dedicated per-customer IPs are not available. See [[concepts/private-networks]]. **Third-party framing (medium confidence).** Vellum calls it an "account-scoped persistent cloud VM" with a browser, filesystem, terminal, connectors, and MCP support. GIGAZINE's launch piece (2026-08-12) says "each agent is controlled by a dedicated cloud PC". **GIGAZINE's wording conflicts with the official docs**, which describe one computer per user that all of that user's Bots share. ## Related Concepts - [[concepts/apps-and-connectors]]: signing in and connectors on this computer - [[concepts/approvals]]: approvals, including local-computer execution, which is separate - [[concepts/private-networks]], [[concepts/proxies]]: network reach and the device-to-computer path - [[concepts/teams-and-enterprises]], [[concepts/enterprise-deployment]] - [[entities/grok-bot]] ## Sources - raw/web_community-computer-and-apps-md.md (https://docs.x.ai/grok-bot/computer-and-apps.md) - raw/web_community-work-md.md (https://cursor.com/docs/grok-bot/work.md) - raw/web_community-computers-md.md (https://cursor.com/docs/grok-bot/computers.md) - raw/web_community-computers-md-2.md (https://docs.x.ai/grok-bot/computers.md) - raw/web_community-security-md.md (https://cursor.com/docs/grok-bot/security.md) - raw/web_community-security-md-2.md (https://docs.x.ai/grok-bot/security.md) - raw/web_community-security-faq-md.md (https://cursor.com/docs/grok-bot/security-faq.md) - raw/web_community-teams-md.md (https://cursor.com/docs/grok-bot/teams.md) - raw/web_community-private-networks-md.md (https://cursor.com/docs/grok-bot/private-networks.md) - raw/web_community-identity-md.md (https://cursor.com/docs/grok-bot/identity.md) - raw/web_community-deployment-md.md (https://cursor.com/docs/grok-bot/deployment.md) - raw/web_community-settings-md.md (https://cursor.com/docs/grok-bot/settings.md) - raw/web_community-troubleshooting-md.md (https://docs.x.ai/grok-bot/troubleshooting.md) - raw/web_community-team-bots-md.md (https://docs.x.ai/grok-bot/team-bots.md) - raw/web_community-official-grok-bot-breakdown-2026.md (https://www.vellum.ai/blog/official-grok-bot-breakdown), third-party - raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md (https://gigazine.net/gsc_news/en/20260812-spacexai-grok-bot/), third-party --- title: "Enterprise Deployment (Installers, MDM, Updates)" type: concept tags: [deployment, mdm, installers, updates, intune, jamf, linux-packages, release-feed] created: 2026-10-21 updated: 2026-10-21 confidence: high sources: ["raw/web_community-deployment-md-2.md", "raw/web_community-deployment-md.md"] --- # Enterprise Deployment (Installers, MDM, Updates) ## Definition How an IT team installs the Grok Bot client apps on member devices with its existing device-management tools, pins versions, and plans updates. Documented on both `docs.x.ai/grok-bot/deployment` and `cursor.com/docs/grok-bot/deployment`; the two pages carry the same installer, feed and path values (the only value difference: the cursor.com page's App Store link is `https://apps.apple.com/app/grok-bot/id6794501026`, the xAI page's is the `/app/id6794501026` form listed below). Turning Grok Bot on and scoping access is a dashboard task covered in [[concepts/teams-and-enterprises]]. ## How It Works | Component | Runs on | Updated by | | --- | --- | --- | | Desktop app | macOS (Apple silicon and Intel), Windows (x64 and Arm64), Linux (x64 and Arm64) | The app, or your package manager for Linux `.deb` and `.rpm` installs | | iPhone app | iOS 18 or later | The App Store | | Android app | Android 9 or later | Google Play | | Hosted computer | The cloud, one per member | Automatically (Cursor's page: "Cursor") | The apps are thin clients for chat, review and approvals; Bots work on the member's hosted [[concepts/cloud-computer]]. MDM reaches the apps and stops there: the hosted computer runs Linux in the cloud and **isn't enrolled in MDM**. To put tooling on hosted computers, Enterprise teams use Team Setup scripts ([[concepts/private-networks]]). **Getting installers:** the **Download Grok Bot** row on the dashboard Grok Bot page (`https://cursor.com/dashboard/bot`); the download page `https://cursor.com/download/bot`; or the release feed for scripted packaging. | Platform | Architectures | Formats | | --- | --- | --- | | macOS | Apple silicon, Intel | `.dmg` | | Windows | x64, Arm64 | Setup `.exe` | | Linux | x64, Arm64 | `.deb`, `.rpm`, AppImage | **Release feed:** `https://api2.cursor.sh/updates/api/download/stable/{platform}/sand`, where `{platform}` is `darwin-arm64`, `darwin-x64`, `win32-x64-user`, `win32-arm64-user`, `linux-x64`, or `linux-arm64`. ```bash curl -s https://api2.cursor.sh/updates/api/download/stable/darwin-arm64/sand ``` Fields: `version`; `commitSha` (Linux download URLs include it); `downloadUrl` (`.dmg` / Setup `.exe` / AppImage); `debUrl`, `rpmUrl` (Linux only). **Versioned URLs** (macOS and Windows): ```text https://downloads.cursor.com/grokbot/stable/darwin-arm64/{version}/Grok_Bot_{version}.dmg https://downloads.cursor.com/grokbot/stable/darwin-x64/{version}/Grok_Bot_{version}_x64.dmg https://downloads.cursor.com/grokbot/stable/win32-x64/{version}/Grok_Bot_{version}_Setup.exe https://downloads.cursor.com/grokbot/stable/win32-arm64/{version}/Grok_Bot_{version}_Setup.exe ``` Linux packages sit under the release's `commitSha`: `https://downloads.cursor.com/grokbot/stable/{commitSha}/linux/x64/` or `.../linux/arm64/` — copy exact file URLs from `downloadUrl`, `debUrl`, `rpmUrl`. **Pushing the app:** - **macOS:** deploy the `.dmg` with Jamf Pro, Kandji, Microsoft Intune or similar. - **Windows:** Intune, Configuration Manager, etc. Silent install: `Grok_Bot_{version}_Setup.exe /S`. Installs **per user** under `%LOCALAPPDATA%\Programs`, so deploy in **user context, not as SYSTEM**; test on a pilot device first. - **Linux:** `.deb`/`.rpm` via Ansible, Puppet, Chef (preferred for managed fleets). - **iPhone:** App Store (`https://apps.apple.com/app/id6794501026`), iOS 18+. - **Android:** Google Play (`https://play.google.com/store/apps/details?id=ai.x.grok.bot`), Android 9+; approve in managed Google Play from Intune or another Android MDM. Members sign in through normal SSO with their Cursor account; no separate login (see [[concepts/identity-and-access]]). ## Key Parameters **Updates:** | Install | How updates arrive | | --- | --- | | macOS `.dmg`, Windows Setup `.exe`, Linux AppImage | App checks automatically and offers each version; **Restart to Update** in settings installs it. No device policy, installer option or dashboard setting turns this off. | | Linux `.deb` / `.rpm` | App doesn't self-update; versions come through your package manager. | **Linux repositories:** the `.deb` adds `/etc/apt/sources.list.d/grok-bot.sources`; the `.rpm` adds `/etc/yum.repos.d/grok-bot.repo`. To control timing, hold the package (`apt-mark hold grok-bot`) or disable the repo (`Enabled: no` in `grok-bot.sources`, or `enabled=0` in `grok-bot.repo`); the package rewrites these files on every install/upgrade but keeps your choice. **Hosted computer image** — cannot be pinned or supplied: | Trigger | Effect | Who | | --- | --- | --- | | New image | Idle computers recreated gradually, no fixed schedule; files carry over, Team Setup reruns | Automatic | | **Update Bot's Computer** | Rebuild on latest image | Member, Settings → Computer → Bot's computer | | **Recreate VMs** | Rebuild on latest image + Team Setup | Enterprise admins (xAI: organization admins) | Running computers pick up manifest changes roughly once a day. Enterprise audit logs record computer operations as `grok_bot_vm` events and multi-member operations as `grok_bot_vm_bulk`. **Version check on a device:** account menu → **Support** → **About** (shows **Version**; **Copy version info**). **Network path:** devices download from `cursor.com` and `downloads.cursor.com`; the app needs the domains in [[concepts/proxies]], including the nested `*.*.cursorvm.com`, allowed and exempt from TLS inspection before you push. ## When To Use Use when packaging Grok Bot for a managed fleet. Pin the version you tested, store the installer in your own repository or MDM package store, and deploy from that copy; record the full Linux URL (it includes `commitSha`). The full ordered checklist is in [[syntheses/enterprise-rollout-guide]]. ## Risks & Pitfalls - **Old builds stop working.** Below the minimum supported version, or past a maximum age when a newer release exists, the app shows a full-window **Update required**. Applies to every install including held Linux packages. Both limits are set remotely and **not published**; nor is how long old installers remain downloadable. Plan a regular update cadence. - **No MSI or PKG** is published; wrap the `.dmg`/`.exe` in your tool's format if required. - **Installing as SYSTEM** on Windows is wrong — it is a per-user installer. - **MDM can't manage the hosted computer**; device-trust IdP policies fail there (see [[concepts/identity-and-access]]). - Members can self-install from the download page unless device policy blocks it; access still depends on plan and **Manage Group Access**. - Cursor's page notes scheduling **Update Bot's Computer** for later "isn't available to every account yet" (as of fetch 2026-10-21). ## Related Concepts - [[concepts/teams-and-enterprises]], [[concepts/proxies]], [[concepts/settings-and-notifications]], [[concepts/mobile-app]] ## Sources - raw/web_community-deployment-md-2.md — https://docs.x.ai/grok-bot/deployment.md - raw/web_community-deployment-md.md — https://cursor.com/docs/grok-bot/deployment.md --- title: "Files and Results" type: concept tags: [grok-bot, attachments, artifacts, workspace, evidence, cloud-agents] created: 2026-10-21 updated: 2026-10-21 confidence: high sources: ["raw/web_community-files-and-results-md.md", "raw/web_community-work-md.md", "raw/web_community-team-bots-md.md", "raw/web_community-computer-and-apps-md.md"] --- # Files and Results ## Definition This page covers how you give a Bot source material (attachments, links, shared files) and how its outputs come back to you: cards in the conversation, files in the shared `/workspace`, and files from Cloud Agents. The docs push for **reviewable** results, meaning artifacts with acceptance criteria and evidence that someone else could check independently. ## How It Works ### Inputs - Use the attachment control, drag files into the composer, or paste images and links. - Supported input types include images, audio, video, PDF and plain text, Word/Excel/PowerPoint, CSV, JSON, YAML, source code, HTML, email files, and Jupyter notebooks. - **iPhone share sheet:** choose **Grok Bot**, pick a chat, then **Attach**. The item is added to that chat's composer. On Android the share sheet currently accepts text (the docs don't say it accepts anything else). - Tell the Bot what each attachment is and how to use it. docs.x.ai `files-and-results.md` and cursor.com `work.md` both give this example: > The PDF is the signed policy. The spreadsheet is this month's transactions. Reconcile the spreadsheet against the policy, cite the relevant policy section for every exception, and return a new spreadsheet plus a short summary. Do not modify the originals. - **Links** work only if the Bot can reach the page from its computer or through a connector. For private pages, sign in through the computer (take over for the login step) or install the relevant connector. Links in messages open in your system browser, and hovering over a link card shows a preview. ### Outputs - Files, images, links, and tool results appear as **cards** in the conversation. Open a card to preview supported formats, then save the file, open the source link, or reply with feedback. - Ask the Bot to **revise the existing artifact** rather than create disconnected copies, for example: "Update the report you just created. Add source links to the first two claims and replace the final table with a CSV attachment." - **Cloud Agent results:** when a Bot hands coding work to a Cloud Agent and the agent finishes, the Bot copies the images, videos, and other files cited in the agent's final report to its computer and attaches them to its reply. On desktop, the Cloud Agent's card shows the same files, and images or videos open full screen. ### The shared workspace - The computer has a shared workspace at `/workspace`. Any of your Bots can read files that another Bot saved there. Use project folders and descriptive names so handoffs are reliable. - **Persistence (cursor.com `work.md`):** files, browser state, and supported sign-ins survive normal computer updates and recovery. Treat temporary directories, manually installed packages, and uncommitted application state as replaceable. A *computer* update (as opposed to a software update) keeps files and logins but removes installed apps and packages. Recover and Reset rebuild from your last saved data, so very recent changes may not come back if the computer couldn't save first. Conversations are stored outside the computer and survive even a reset. - The workspace is fine for intermediate material, but the **conversation should still contain the final result** or a clear link to it. - **Team Bots:** files you upload in your own chat with a Team Bot land on **your** computer, not the owner's. Team reference files (the **Files** setup card) accept .txt, .md, .markdown, .csv, .json, .yaml, and .yml, up to 256,000 characters per file (see [[concepts/team-bots]]). ## Key Parameters | Limit | Value | |---|---| | Attachments per message (desktop composer) | 6 | | Documents, images, audio | up to 25 MB each | | Video | up to 200 MB | | Shared workspace path | `/workspace` | | Android share sheet | accepts text (as of fetch 2026-10-21) | | Retention of generated files | not documented beyond the update, recover, and reset behavior above | **Reviewable output formats to request:** a document with headings and source links, a spreadsheet with defined columns and formulas, a slide deck with speaker notes, a folder of screenshots and logs, an unsent draft message, or a short recommendation followed by its evidence. **For consequential work, ask the Bot to separate:** (1) facts found in source systems, (2) assumptions or inferences, (3) actions already completed, (4) actions waiting for approval, (5) unresolved questions. **Evidence to ask for:** direct source links, screenshots with the relevant state visible, timestamps and time zones, input and output file names, a concise action log, and an explicit list of anything the Bot couldn't verify. ## When To Use - Attach files when the Bot needs your exact source of truth (a policy, an export). - Use links when the live system is authoritative and the Bot can reach it. - Use `/workspace` for multi-Bot pipelines and intermediate files. Put the final deliverable in the conversation. - Ask for an evidence package whenever a decision or external action depends on the result. ## Risks & Pitfalls - Large, encrypted, damaged, or unusual files may not be readable. Export them as PDF, CSV, plain text, or an image. - Don't rely on a screenshot alone for fast-changing data. Keep a link or an export from the source system. - Check a link's destination before entering credentials or approving an external action. - Files in `/workspace` are visible to **all** your Bots and are not isolated per Bot (see [[concepts/cloud-computer]]). Deleting a Bot removes its profile, conversation, and routines, but files and sign-ins on the shared computer can remain and need separate cleanup (cursor.com `work.md`). - Temporary directories and installed packages can disappear after a computer update or recovery. Copy important results into `/workspace` or attach them to the conversation. ## Related Concepts - [[concepts/messaging-and-collaboration]] - [[concepts/cloud-computer]] - [[concepts/apps-and-connectors]] - [[concepts/privacy-and-data]] - [[concepts/teams-and-enterprises]]: Cloud Agent delegation - [[concepts/mobile-app]] ## Sources - `raw/web_community-files-and-results-md.md`: https://docs.x.ai/grok-bot/files-and-results.md - `raw/web_community-work-md.md`: https://cursor.com/docs/grok-bot/work.md - `raw/web_community-team-bots-md.md`: https://docs.x.ai/grok-bot/team-bots.md - `raw/web_community-computer-and-apps-md.md`: https://docs.x.ai/grok-bot/computer-and-apps.md (same persistence guidance on docs.x.ai) --- title: "Identity and Access (SSO, SCIM, Okta, Entra ID)" type: concept tags: [identity, sso, saml, scim, okta, entra-id, fastpass, conditional-access, bot-identity] created: 2026-10-21 updated: 2026-10-21 confidence: high sources: ["raw/web_community-identity-and-access-md.md", "raw/web_community-identity-md.md", "raw/web_community-security-md.md", "raw/web_community-security-md-2.md"] --- # Identity and Access (SSO, SCIM, Okta, Entra ID) ## Definition How people get into Grok Bot and how Bots act under their identity. Grok Bot uses the member's **Cursor account**, so there is **no separate Grok Bot app** in Okta or Entra ID — everything happens on the Cursor app you already have. Documented on `docs.x.ai/grok-bot/identity-and-access` and `cursor.com/docs/grok-bot/identity` (same content; Cursor says "connected-app sign-in" where xAI says "plugin sign-in"). ## How It Works **Sign-in.** Cursor SSO is **SAML 2.0** and works with Okta, Microsoft Entra, Google Workspace and OneLogin; SSO can be required for all members (which blocks password login). Access requires membership of your Cursor team. With auto-provisioning on, a user joins at first sign-in; otherwise provision them, then have them sign in to Cursor for a seat or invite them from the dashboard. **SCIM is Enterprise only.** Provisioning is SCIM 2.0 and deprovisioning is automatic: removing the user in the IdP removes them from Cursor. **Bot identity** (security pages, both hosts). A Bot has no identity or credentials of its own: - Bots act as the signed-in member, never hold more access than that person, and every action is attributable to a named member. Team-managed connectors are the one exception (may use team or service-account credentials). - Connector OAuth tokens stay on Cursor's backend and are never stored on the computer. - For login, two-factor and payment steps the Bot hands the computer to the member; a masked secret request keeps values out of the transcript and away from the model. **Two separate changes** an Okta/Entra admin makes: 1. **Assign the Cursor app** — controls Grok Bot sign-in for people outside engineering. 2. **Add an authentication rule** — lets members sign in to IdP-provisioned apps *inside the hosted computer's browser*. This never blocks Grok Bot sign-in and doesn't apply to plugin/connected-app sign-in, which doesn't go through the computer. ## Key Parameters ### Assign the Cursor app If Cursor is assigned only to engineering, others fail with **User is not assigned to this application**, or their invite never completes. Widen assignment; don't create a second app. **Okta:** **Admin Console → Applications → Applications** → your Cursor app → **Assignments** → **Assign → Assign to Groups** (or **Assign to People**) → add every group that should get Grok Bot. With SCIM, assign the same groups to the Cursor SCIM app and push them (users appear only after SCIM assignment). Confirm each group is also on the SAML app — SCIM without SSO assignment still blocks first sign-in. With org-level identity, map the directory group to the team after sync; use separate groups for app assignment and Group Push. **Entra ID:** **Microsoft Entra admin center → Enterprise applications** → your Cursor app → **Users and groups** → **Add user/group**. With SCIM, assign the same groups to the Cursor provisioning app, turn provisioning on, scope to assigned users and groups. Confirm the assignment grants SSO, not report-only. Group-based assignment requires **Entra ID P1 or P2**; nested groups are not included. ### Allow sign-in to IdP apps from the computer The hosted computer runs Linux, isn't enrolled in MDM by default, and can't run device-trust agents such as **Okta FastPass**. Any rule requiring FastPass, a registered/managed/compliant device, or a phishing-resistant factor only FastPass can satisfy fails in the computer browser. Don't disable these company-wide; add a higher-priority scoped rule. Working methods: password plus a remote-friendly second factor (Okta Verify push, authenticator app), or passkeys in a password manager installed via a Team Setup script (Enterprise only). **Okta (Identity Engine):** the computer matches device platform **Other Desktop** (no Linux checkbox). Per app the Bot opens: 1. **Admin Console → Security → Authentication Policies**, open the app's policy (find it via **Applications** → app → **Sign On**). 2. Add a rule **above** the FastPass, managed-device and deny catch-all rules, e.g. named **Grok Bot computer (Linux)**. 3. IF: scope to a group (the Cursor-assigned group works). 4. **Device platform** = **Other Desktop**, **Device state** = **Any**; don't require **Registered**, **Managed**, or FastPass-dependent device assurance. 5. THEN: **Allowed after successful authentication** with **Password + Another factor**. 6. Save; confirm it sits above the unmanaged-device deny catch-all. Classic Engine: allow **Other Desktop** without requiring **Device Trust = Trusted**. **Entra ID (Conditional Access):** blocking grants are compliant device, hybrid joined device, a phishing-resistant strength only a platform/managed passkey meets, and approved client app / app protection policy. 1. **Microsoft Entra admin center → Protection → Conditional Access → Policies**. 2. Find policies applying those grants to the needed apps; leave them on. 3. Add a higher-priority policy for Grok Bot users on Linux (or exclude them). 4. Users = Cursor-assigned group; target only the needed apps (**All resources** only if you accept that scope). 5. **Device platforms**: include **Linux**, exclude **Windows** and **macOS**. 6. Grant: **Require multifactor authentication** only. 7. Start in report-only mode, then turn on. 8. On existing compliant-device policies, exclude the group or Linux. ## When To Use During rollout, before members start (step 3 of the deployment checklist in [[concepts/enterprise-deployment]]). Requiring managed devices for Grok Bot *sign-in itself* still works, because a device-aware IdP policy gates the member's device, not the hosted computer. Success checks: a non-engineering user completes Cursor SSO without the "not assigned" error; IdP apps open from the computer without FastPass/compliance errors; laptop sign-ins are unchanged. ## Risks & Pitfalls - **FastPass can't run on Linux**, so the computer can never satisfy FastPass/managed-device rules — the most common cause of in-computer sign-in failures. - **SCIM assigned but SAML not** blocks first sign-in. - **Unassigning** blocks SSO sign-in; with SCIM it also removes the user from Cursor, without SCIM remove them in the dashboard. - **Terminating a computer doesn't remove access** (Enterprise, organization admin): remove the member from the team or turn off Grok Bot for their group, and revoke their sessions in your identity provider. Revoking the user in your IdP ends their in-computer app sessions. - All of a member's Bots share one computer, so a login there is available to every Bot that member runs ([[concepts/cloud-computer]]). - Deleting a Bot doesn't remove computer files or browser sessions; clean up `/workspace` and sign out of sites. ## Related Concepts - [[concepts/teams-and-enterprises]], [[concepts/private-networks]], [[concepts/approvals]], [[syntheses/security-model]], [[concepts/apps-and-connectors]] ## Sources - raw/web_community-identity-and-access-md.md — https://docs.x.ai/grok-bot/identity-and-access.md - raw/web_community-identity-md.md — https://cursor.com/docs/grok-bot/identity.md - raw/web_community-security-md.md — https://cursor.com/docs/grok-bot/security.md (Identity and sign-ins section) - raw/web_community-security-md-2.md — https://docs.x.ai/grok-bot/security.md (Identity and sign-ins section) --- title: "Inference Hooks" type: concept tags: [inference-hooks, dlp, security-vendor, webhooks, enterprise, beta, audit-log] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-inference-hooks-md.md", "raw/web_community-security-md.md", "raw/web_community-security-md-2.md", "raw/web_community-security-faq-md.md", "raw/web_community-teams-md.md", "raw/web_community-security-faq-md-2.md"] --- # Inference Hooks ## Definition Inference hooks send every governed Grok Bot **model request**, signed, to an AI security server that your organization runs or uses through a security vendor. Cursor waits for an `allow` or `deny` verdict and stops the request on a deny. This lets your own policy check prompts and the content Grok Bot gathers (web pages, app results) before a model sees them. As of fetch 2026-10-21, the feature is in **beta for Enterprise teams**, enabled through your account manager. It covers **Grok Bot only**. **Host contradiction (unresolved in the sources).** In this corpus only cursor.com pages document inference hooks (`inference-hooks.md`, `security.md`, `security-faq.md`, `teams.md`). cursor.com `security.md` says to use inference hooks to check what reaches the model "against your own data loss prevention policy", lists **Inference Hooks** among its Enterprise-only dashboard items, and its security FAQ answers "Can our own security tooling check what Grok Bot sends to models?" with "Yes, with inference hooks, an Enterprise beta." docs.x.ai `security.md` says, under its network-policy bullet, "Dedicated data loss prevention hooks are not available", does not list Inference Hooks in its Enterprise-only list, and its security FAQ has no inference-hook question. The sources do not say which statement reflects the current product, so confirm availability with your account team. ## How It Works - Before each model call, Cursor sends a signed `prompt` frame containing the whole conversation so far, and holds the request. A turn with several model calls sends one frame per call. - `allow` lets the request proceed. `deny` stops it, ends the turn, and shows the sender a block notice. - **Only model inputs are checked.** A tool call and its result reach your endpoint only after the tool has run. A Bot's reply to the member is itself a tool call, so the member sees the reply before any frame contains it. A deny stops the *next* model call, not an action that already happened. - A failure **never counts as a deny**. The mode decides what happens on a failure. - **Enforce Verdicts** turns sending on and off. While it is off, Cursor sends nothing and keeps your settings. The mode applies only while it is on. **Setup** (team admin on an Enterprise team with hooks enabled): 1. Prepare a vendor integration or your own server. The **Security Vendor** list has presets. 2. In the dashboard, open **Inference Hooks** under **Security & Compliance** on the Grok Bot page. 3. Click **Set Up Endpoint**, choose your vendor or **Custom Endpoint**, and enter the **Endpoint URL** plus any **Custom Headers**. 4. Copy the signing secret (starts with `whsec_`). It is shown once. 5. Click **Test Connection**. 6. New configs start in **Shadow Mode**. Turn on **Enforce Verdicts** to start sending. Once verdicts look right, set **Mode** under **Failure Handling** to **Allow the Request** or **Block the Request**. Settings changes take effect within about a minute. While hooks are active, Grok Bot skips some pre-turn model calls, so replies can start more slowly. ## Key Parameters | Mode | On deny | When the check fails | |---|---|---| | **Shadow Mode** (default) | Lets the request through | Lets the request through | | **Allow the Request** | Blocks | Proceeds uninspected (fails open) | | **Block the Request** | Blocks | Blocks (fails closed) | In Shadow mode the frames are sent in the background, members see nothing, and denials and failures are audit-logged. Under heavy load a Shadow check can be skipped, and then no audit row is written. - **Endpoint:** must be an `https://` URL on port 443, on a publicly routable host with a public-CA certificate. Private, loopback, and CGNAT addresses, redirects, and reverse tunnels are refused. The URL can't contain credentials, a query string, or a fragment, and is limited to 2,048 characters. - **Custom headers:** up to 16. Names up to 128 characters (letters, digits, hyphens, no underscores). Values are printable ASCII up to 8,192 characters and are write-only. Changing the URL clears saved values. Reserved names include `Host`, `Cookie`, `User-Agent`, and any name starting with `Content-`, `Proxy-`, `X-Forwarded-`, `Webhook-`, or `X-Cursor-`. - **Signing:** uses Standard Webhooks headers `webhook-id` (equal to `request_id`), `webhook-timestamp`, and `webhook-signature`. The signature is `v1,` followed by the base64 HMAC-SHA256 of `{webhook-id}.{webhook-timestamp}.{raw body}`, keyed with the base64-decoded part after `whsec_`. **Rotate Secret** takes effect within about a minute. - **Verdict Timeout (ms):** 1 to 10,000, default **5,000**. On a connection failure Cursor retries once, with the same `webhook-id`. - **Requests Inspected (%):** sticky per turn. Requests outside the share proceed uninspected even in Block mode. 0% turns inspection off. - **Size limits:** responses are read up to 64 KiB, uncompressed. A frame over 64 MiB counts as a failure. - **Blocked Message:** up to 500 characters. Default "Your organization's security policy blocked this request." `deny_reason` is truncated to 500 characters. **Request.** An HTTPS `POST` with `user-agent: cursor-inference-hooks/1`. The JSON fields are `type` (always `prompt`), `request_id`, `tenant_id`, `actor` (`type` `user`, an opaque `id`, and `email_address` always null), `source.application` (`grok-bot` or `config-test`), `session_id` (per Bot, not per conversation), `model` (for reporting, not allowlisting), `messages`, and `metadata` (always empty). Message blocks are `text`, `tool_use`, `tool_result`, and `attachment`. Images arrive without their bytes. Secret values are never included. System prompts, tool definitions, and hidden reasoning are excluded. **Verdict.** HTTP 200 with a body like: ```json { "action": "deny", "deny_reason": "...", "reference_id": "policy-7f3c21" } ``` `reference_id` can be up to 50 characters from `A-Z`, `a-z`, `0-9`, `.`, `_`, `:`, `/`, and `-`. Unknown fields are ignored. **Circuit breaker.** Sustained failures your server is responsible for (timeouts, error statuses, TLS errors, malformed verdicts) stop sending, and then your mode applies to every covered request: Block mode blocks them all, while Allow and Shadow let them through uninspected. Recovery probes start 10 minutes after the trip. *Any* settings change, including a secret rotation, stops automatic recovery. Use **Reset** instead. ## When To Use Use inference hooks to apply your own DLP or AI-security policy to what Grok Bot sends to models. They are not an action control. Keep the network policy and Auto Review in place for actions (see [[concepts/approvals]]). ## Risks & Pitfalls - **Not covered:** model responses, ancillary calls (chat titles, Auto Review, memory saves), Cloud Agents, editor agents, the CLI, the Agent SDK, Tab, embeddings, and local models. Actions are not checked as they run. - **One block can poison a conversation.** The blocked message is resent with every later request, so a whole-transcript policy keeps denying. The fix is **Edit**, **Remove blocked message** (Slack), or **Reset**. If it blocks again after a reset, the matching content is likely in the Bot's memory or files. - **Block mode** shows **Your organization's policy check is unavailable** on any failed check, including Cursor-side causes. If Cursor can't load your settings at all, the request proceeds as failed-open with `config_unavailable`. While Cursor has paused inference hooks (an emergency pause), no frames are sent and the audit log records most covered requests as failed open with reason `kill_switch`. - **Shadow mode still sends conversations to your endpoint.** Tell members before you enable **Enforce Verdicts**. - Verify the signature. Don't trust the User-Agent or source IP. Scan every block type, not just `text`. Don't rely on `tool_result.is_error`. - **Audit events:** `inference_hooks_config_updated`, `inference_hooks_signing_secret_generated`, `inference_hooks_request_denied`, `inference_hooks_request_failed_open`, `inference_hooks_request_failed_closed`, and `inference_hooks_circuit_breaker_tripped`. Allowed requests write no event. - **Privacy:** Cursor discards frames after the verdict. Your endpoint's retention is governed by your agreement with your vendor. Hooks run regardless of Privacy Mode. See [[concepts/privacy-and-data]]. ## Related Concepts - [[syntheses/security-model]], [[concepts/approvals]], [[concepts/privacy-and-data]] - [[concepts/teams-and-enterprises]], [[syntheses/troubleshooting]] ## Sources - raw/web_community-inference-hooks-md.md (https://cursor.com/docs/grok-bot/inference-hooks.md) - raw/web_community-security-md.md (https://cursor.com/docs/grok-bot/security.md) - raw/web_community-security-md-2.md (https://docs.x.ai/grok-bot/security.md) - raw/web_community-security-faq-md.md (https://cursor.com/docs/grok-bot/security-faq.md) - raw/web_community-teams-md.md (https://cursor.com/docs/grok-bot/teams.md) - raw/web_community-security-faq-md-2.md (https://docs.x.ai/grok-bot/security-faq.md) --- title: "Messaging and Collaboration" type: concept tags: [grok-bot, chat, group-chat, handoffs, voice, shortcuts] created: 2026-10-21 updated: 2026-10-21 confidence: high sources: ["raw/web_community-chat-and-collaboration-md.md", "raw/web_community-work-md.md", "raw/web_community-mobile-md.md"] --- # Messaging and Collaboration ## Definition Working with Grok Bot is meant to feel like messaging a teammate. You talk to a Bot one-to-one, put several Bots in a **group chat**, or let Bots **hand work to each other** asynchronously. The docs' standing advice is to keep requests natural but make two things explicit: the outcome you want and the decision boundary (what the Bot may change and where it must stop for you). ## How It Works ### In a conversation you can - Paste text, links, and images, and attach local files (see [[concepts/files-and-results]]) - Reference a saved skill with `/`, and mention a Bot, group, routine, or connector/app with `@` - Reply to a specific message or react to it - Send another instruction while work is in progress - Dictate, start a live voice chat, or play a voice memo a Bot sent. These are on docs.x.ai `chat-and-collaboration.md` and `mobile.md`; cursor.com `work.md` doesn't list voice features. The transcript shows tool activity, computer use, created files, questions, approval requests, and voice memos alongside ordinary messages. **Voice.** - **Dictate:** on desktop choose **Start voice input** or press `Cmd/Ctrl+D` while the prompt is focused. On mobile use **Start dictation**. The text lands in the composer for you to edit before sending. - **Voice memo:** **Play voice memo** or **Pause voice memo**. Expand the memo to read its transcript. - **Voice chat:** **Start voice chat** appears when the composer is empty. While a call is live the app shows **Voice chat**, and the desktop sidebar can show **Voice chat in progress**. A **Voice chat** card may appear afterwards. There is no keybinding. **Drafts.** When a Bot prepares an email or Slack message, an editable draft appears. Review the recipients and body, then choose **Send** (or **Send message** for a Slack draft on desktop), or choose **Discard**. **Redirecting.** A direct message from you takes priority over background work and can redirect the current turn. Send "Stop now" to end work immediately. **Stopping doesn't undo actions the Bot already completed.** ### Group chats 1. Choose **New** / **New chat** in the sidebar, then select **two to six Bots**. On mobile use **+ → New Group Chat**. 2. Edit the generated group name if you like. Membership can be changed later. 3. Describe the shared outcome and who owns the next step. How to direct messages in a group: - Write normally and the Bots decide who responds. - Type `@` and pick a Bot when one teammate owns the request. - Choose **Reply** on a Bot's message to send your answer to that Bot. - Mention several Bots only when each one is needed. - Use `@everyone` sparingly, for group-wide updates. Example kickoff from both hosts: > @Researcher gather the source material and link every claim. @Writer turn the findings into a launch draft. @Reviewer check the draft against the sources and list only blocking issues. Do not publish anything. What appears in a group: your messages can carry attachments. Bot replies can include files, questions, and forms, and your answer goes back to the Bot that asked. **Approval requests, secret and login requests, and email or Slack drafts do not appear in a group.** Bot-to-group handoff messages are text-only (docs.x.ai says "currently"), so a Bot should send an image **directly** to a teammate Bot that needs to inspect it. ### Bot-to-Bot handoffs A Bot can send an asynchronous message to another Bot. The receiving Bot wakes, handles the request, may reply later, and you can see the handoff in the conversation. Handoffs suit cases where one Bot owns a source system and another owns the deliverable, where a specialist should review a draft, where a blocker belongs to another role, or where a long job should continue without you coordinating each step. cursor.com adds that a handoff needs a Bot acting for one person. From the app or your own Slack DM any Bot can hand off. From a Slack channel thread only a [[concepts/team-bots]] Team Bot can, on behalf of the person who mentioned it. ### Threads, reactions, search - Reply **in a thread** when feedback concerns one result or one approval request. - Reactions are for lightweight acknowledgement only. A reaction alone should not carry a safety-critical decision. - Search and the command palette switch between Bots and groups and find prior messages, files, links, and routines. Mobile search filters are **Messages**, **Bots**, **Group Chats**, **Files**, **Routines**, and **All**. Search availability can vary during rollout. If cross-conversation results aren't available, use the Bot's own conversation history. - Mobile saves drafts per conversation when you navigate away. ## Key Parameters | Item | Value | |---|---| | Bots per group chat | 2 to 6 | | Search palette | `Cmd/Ctrl+K` | | Search Bots | `Cmd/Ctrl+Shift+F` | | New Bot / new chat | `Cmd/Ctrl+N` | | Find in chat | `Cmd/Ctrl+F` | | Compact sidebar | `Cmd/Ctrl+B` | | Focus prompt | `Cmd/Ctrl+I` | | Add selected text to chat | `Cmd/Ctrl+L` | | Sidebar Bot 1–9 | `Cmd/Ctrl+1` … `Cmd/Ctrl+9` | | Previous / next Bot | `Alt+↑` / `Alt+↓` | | Back / forward | `Cmd/Ctrl+[` / `Cmd/Ctrl+]` | | Cycle Bots | `Control+Tab` / `Control+Shift+Tab` (always Control, including on macOS) | | Connect apps | `Cmd/Ctrl+Shift+M` or `Cmd/Ctrl+Shift+W` | | Toggle Bot settings | `Cmd/Ctrl+Shift+,` | | Conversation details | `Cmd+Shift+I` or `Cmd+Alt+B` (macOS); `Ctrl+Alt+B` (Windows/Linux) | | Send / new line | `Enter` (or `Cmd/Ctrl+Enter` from anywhere) / `Shift+Enter` | | Dictate | `Cmd/Ctrl+D` | | Settings | `Cmd/Ctrl+,` | | Zoom in / out / reset | `Cmd/Ctrl+=` / `Cmd/Ctrl+-` / `Cmd/Ctrl+0` | | Fullscreen | `F11`; on macOS also `Control+Cmd+F` | (The shortcut list comes from docs.x.ai `chat-and-collaboration.md`.) ## When To Use - **One-to-one chat:** a single owner and outcome. - **Group chat:** several Bots share one outcome and the handoffs themselves should be visible. Assign one owner per stage. - **Handoffs:** long-running or cross-role work. Ask for a single owner at each stage. - **New thread or new Bot:** when the conversation has drifted into a different long-lived job. ## Risks & Pitfalls - **Parallel handoffs** create duplicate work and noisy updates. - **Groups don't show approvals, secrets, logins, or drafts.** Handle those in a 1:1 conversation. - **"Stop now" doesn't roll back** completed actions. - **Images don't travel in Bot-to-group handoffs.** Send images directly to the Bot that needs them. - **Reactions as approvals:** don't use a reaction for a safety-critical decision. Write the instruction. - **Slack channel handoffs** work only from a Team Bot. ## Related Concepts - [[concepts/bots]] - [[concepts/team-bots]] - [[concepts/files-and-results]] - [[concepts/approvals]] - [[concepts/mobile-app]] - [[concepts/skills-routines-and-automations]] ## Sources - `raw/web_community-chat-and-collaboration-md.md`: https://docs.x.ai/grok-bot/chat-and-collaboration.md - `raw/web_community-work-md.md`: https://cursor.com/docs/grok-bot/work.md - `raw/web_community-mobile-md.md`: https://docs.x.ai/grok-bot/mobile.md --- title: "Grok Bot Mobile App" type: concept tags: [grok-bot, mobile, ios, android, ipad, notifications] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-mobile-md.md", "raw/web_community-get-started-md.md", "raw/web_community-get-started-md-2.md", "raw/web_community-overview-md.md", "raw/web_community-faq-md.md", "raw/web_community-deployment-md.md", "raw/web_community-deployment-md-2.md", "raw/web_community-settings-and-notifications-md.md", "raw/web_community-grok-bot-md.md", "raw/web_community-approvals-security-and-privacy-md.md", "raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md"] --- # Grok Bot Mobile App ## Definition The Grok Bot mobile app is a client for iPhone, iPad and Android. You use it to start work, answer Bots' questions, approve steps and review results away from your desk. It connects to the same Bots, conversations, routines, connectors and shared cloud computer as the desktop app. Work keeps running in the cloud when the app is closed (x.ai mobile page). ## How It Works ### Requirements and install | Item | Value | |---|---| | iPhone | iOS 18 or later. Install from the [App Store](https://apps.apple.com/app/id6794501026) | | iPad | iPadOS 18 or later (runs the iOS app) | | Android | Android 9 or later. Install from [Google Play](https://play.google.com/store/apps/details?id=ai.x.grok.bot) (package `ai.x.grok.bot`) | | Plan | The same eligible plans as desktop. See [[concepts/plans-and-access]] | | Network | Internet connection | Managed devices: both deployment pages say to push Android through managed Google Play from Intune or another Android device management tool. iPhone users install from the App Store ([[concepts/enterprise-deployment]]). ### Sign in and set up 1. Open Grok Bot and choose **Log In or Sign Up**. 2. Complete Cursor authentication in the browser. 3. Return to the app. If your access comes from SuperGrok, choose **Link Grok Account**, finish linking in the browser, then choose **Finished Linking? Refresh My Status**. 4. **New users** take the first-run tour, choose a first Bot and wait while the shared computer is set up. **Existing users** go straight to their synced Bot list. The Cursor get-started page adds that access from your plan or SuperGrok link "carries over without a second purchase". It also links a "Subscribe from the mobile app" help page for accounts without access. ### What you can do on mobile - **Message a Bot:** - send text - **Start dictation** - **Start voice chat**, used when the composer is empty. While the call is live, a **Voice chat** control stays under the header - play a **voice memo** - take or attach a photo, or attach a file - mention another Bot or `@everyone` in a group - reply in a thread, or react to a message - **Drafts:** they are saved per conversation. When a Bot prepares an email or Slack draft, choose **Send** or **Discard**. - **Share sheet:** - On iPhone, choose **Grok Bot**, pick a chat, then **Attach**. This works for photos, files, links and text. - On Android, the share sheet "currently accepts text". - **Bots and groups:** the **+** control → **New Bot** or **New Group Chat**. You can also edit profiles, manage members, pin or hide conversations, and delete a Bot. - **Computer:** - Watch the work, or inspect the current screen. - Take over for a password, 2FA code or CAPTCHA, then hand back with **I'm done**. - It is the same shared computer every Bot uses. - **Approvals:** **Allow once**, **Always allow** and **Deny** work the same on iPhone as on desktop. - **Routines:** in a Bot's profile, view the schedule, next run, instruction and **Run history**, use **Active** to pause or resume, or delete the routine. - **Search:** filter by **Messages**, **Bots**, **Group Chats**, **Files**, **Routines** or **All**. - **Settings:** - account - plugins - Bot settings - Auto Review (when available) - appearance - **Language**: **System** follows the device, and the list is shorter than on desktop - usage, or an App Store or Google Play subscription - sign out or delete your account ## Key Parameters | Parameter | Value | |---|---| | Minimum OS | iOS 18 / iPadOS 18 / Android 9 | | Desktop-only actions | Editing a routine's schedule or instruction, and testing routines. Teaching a workflow by demonstration. "Some advanced desktop controls". The Cursor page also lists "updating the computer" | | Notification routing | While you use the desktop app on macOS or Windows, notifications from Bots you own go to the desktop. When a Bot needs input, your phone is notified about 3 minutes later unless you act first. Finished replies skip the phone. On Linux, the phone is notified as usual | | Notification permission | Requested during first run. Both the device permission and the Bot's **Notifications** setting must allow it | ## When To Use Use mobile to approve or deny actions, unblock a Bot with a takeover, start quick tasks by voice or share sheet, and pause routines. Use desktop to build or test routines, teach workflows and use advanced settings ([[concepts/skills-routines-and-automations]], [[concepts/settings-and-notifications]]). ## Risks & Pitfalls - **Android support differs by source and date.** - x.ai docs: the x.ai mobile, FAQ, get-started and overview pages list Android (Android 9 or later on the mobile and FAQ pages). - Cursor docs: the Cursor deployment page also lists Android 9 or later, but the Cursor `grok-bot.md` overview page says the mobile app runs "on iPhone", and the Cursor get-started page covers iOS only. - At launch, Gigazine (2026-08-12) reported that an Android version was not available. - Treat Android as available per the x.ai docs fetched 2026-10-21. - **The Android share sheet is text-only** for now. - **Missing phone notifications** often come from routing: the desktop app is "in use". Check that before you suspect a bug. - **Taking over on mobile acts on the shared computer.** Anything you sign into there is available to every Bot on your account. ## Related Concepts [[syntheses/getting-started]], [[concepts/messaging-and-collaboration]], [[concepts/approvals]], [[concepts/cloud-computer]], [[concepts/settings-and-notifications]], [[concepts/plans-and-access]] ## Sources - raw/web_community-mobile-md.md — https://docs.x.ai/grok-bot/mobile.md - raw/web_community-get-started-md.md — https://cursor.com/docs/grok-bot/get-started.md - raw/web_community-get-started-md-2.md — https://docs.x.ai/grok-bot/get-started.md - raw/web_community-overview-md.md — https://docs.x.ai/grok-bot/overview.md - raw/web_community-faq-md.md — https://docs.x.ai/grok-bot/faq.md - raw/web_community-deployment-md.md — https://cursor.com/docs/grok-bot/deployment.md - raw/web_community-deployment-md-2.md — https://docs.x.ai/grok-bot/deployment.md - raw/web_community-settings-and-notifications-md.md — https://docs.x.ai/grok-bot/settings-and-notifications.md - raw/web_community-grok-bot-md.md — https://cursor.com/docs/grok-bot.md - raw/web_community-approvals-security-and-privacy-md.md — https://docs.x.ai/grok-bot/approvals-security-and-privacy.md - raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md — https://gigazine.net/gsc_news/en/20260812-spacexai-grok-bot/ (third-party) --- title: "Plans and Access" type: concept tags: [grok-bot, plans, pricing, usage, supergrok, cursor, enterprise] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md", "raw/web_community-overview-md.md", "raw/web_community-grok-bot-md.md", "raw/web_community-get-started-md-2.md", "raw/web_community-get-started-md.md", "raw/web_community-faq-md.md", "raw/web_community-teams-and-enterprises-md.md", "raw/web_community-teams-md.md", "raw/web_community-settings-and-notifications-md.md", "raw/web_community-team-bots-md.md", "raw/web_community-tag-on-x-md.md", "raw/web_community-troubleshooting-md.md", "raw/web_community-mobile-md.md", "raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md", "raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md", "raw/web_community-official-grok-bot-breakdown-2026.md"] --- # Plans and Access ## Definition The official docs describe no standalone Grok Bot price or plan. Access comes as an inclusion in a Cursor or SuperGrok subscription (and, per Cursor docs, a Grok Business seat), with its **own weekly usage allowance**. (The docs do mention subscribing from the mobile app for accounts without access, and the x.ai FAQ refers to "Grok Bot subscriptions"; neither gives a price or plan name.) The docs call the "Plans and billing" help page (cursor.com/help/grok-bot/plans) "the canonical plan and usage matrix". That page is not in this corpus. ## How It Works ### Which plans include Grok Bot, by source and date | Plan | Launch, 2026-08-11 (per Gigazine, 2026-08-12) | x.ai news, 2026-08-26 | x.ai docs (fetched 2026-10-21) | Cursor docs (fetched 2026-10-21) | |---|---|---|---|---| | SuperGrok | — | Yes | Yes (link an individual subscription) | Yes (individual SuperGrok link) | | SuperGrok Plus | — | Yes | Yes | Yes | | SuperGrok Heavy | Yes | Yes | Yes | Yes | | Cursor Pro | — | Yes | "every paid individual Cursor plan" | "every paid individual Cursor plan" | | Cursor Pro+ | — | Yes | (as above) | (as above) | | Cursor Ultra | Yes | Yes | (as above) | (as above) | | Cursor Teams Standard | — | Yes | Yes ("Self-serve Cursor Teams Standard and Premium seats include Grok Bot", FAQ) | Yes (Teams row: "Included; every member has access") | | Cursor Teams Premium | Yes | Yes | Yes | Yes | | Self-serve Grok Business seat | — | not mentioned | **not mentioned** | **Yes** (overview, get-started, teams page) | | Enterprise | Waitlist (Gigazine) | "Enterprise users can join a waitlist" | FAQ: "Enterprise access is rolling out", contact your Cursor account team. Teams page: "Enable Grok Bot for your team from your dashboard" | Teams page: "Enable Grok Bot for your team from your dashboard" | "—" means that source does not list the plan as included at that time. ### How usage works - **Separate allowance.** "Grok Bot comes with its own usage, separate from your Grok and Cursor plans" (x.ai news, 2026-08-26). Work you hand to a Bot doesn't count against your existing usage. - **Weekly reset.** "Usage resets weekly" (both overviews). Grok Bot subscriptions include weekly usage, and eligible accounts can add **on-demand usage billed from model and token cost** (x.ai FAQ). - **Two subscriptions.** If you have both a Cursor and a SuperGrok subscription, "Grok Bot uses whichever has more usage" (x.ai FAQ). - **Teams.** Every member has access, and "usage follows the seat's allowance" (both teams pages). - **Team Bots.** A teammate's chats with a Team Bot count against that teammate's own allowance, not the owner's. A routine counts against the allowance of the person who set it up. Slack messages that don't come from a linked teammate count against the owner. - **Tag @bot on X.** Tagged requests "use your Grok Bot usage". - **Where to see usage.** **Settings → Usage & Billing** shows weekly included and on-demand usage, and the account menu can show **Weekly usage**. On mobile, Settings can "manage an eligible App Store or Google Play subscription". - **No dedicated spend cap.** "A separate Grok Bot spend cap is not available today". Account-level on-demand controls apply instead (both teams pages). ### Getting access from SuperGrok 1. Sign in with your Cursor account. 2. Link the Grok account: - On desktop: **Settings → Usage & Billing** → **Get Access with Grok**. - On iPhone and Android: **Link Grok Account**, then **Finished Linking? Refresh My Status**. ## Key Parameters | Parameter | Value | Source | |---|---|---| | Usage period | Weekly reset | x.ai and Cursor overviews | | On-demand | For "eligible accounts", billed from model and token cost | x.ai FAQ | | Dual subscription rule | The subscription with more usage is used | x.ai FAQ | | Grok Bot-specific spend cap | Not available today | x.ai and Cursor teams pages | | Enterprise | Waitlist (2026-08-26) → "rolling out" / enable from dashboard (as of fetch 2026-10-21) | x.ai news; x.ai FAQ; teams pages | | Standalone price | **Not documented officially.** Vellum's breakdown claims a 14-day free trial, then $200 per month | Vellum (third-party) | | Launch plan prices | SuperGrok Heavy $300/month, Cursor Ultra $200/month, Cursor Teams Premium $120 per user per month | Gigazine (third-party) | ## When To Use - To answer "do I already have Grok Bot?": any paid individual Cursor plan, Cursor Teams, or a SuperGrok-tier link qualifies (as of fetch 2026-10-21). - To explain why a Bot stopped: check whether weekly usage is exhausted or an on-demand spending limit was reached. Troubleshooting points to the account access page or **Usage & Billing**. - To plan an enterprise rollout: see [[concepts/teams-and-enterprises]] and [[syntheses/enterprise-rollout-guide]]. ## Risks & Pitfalls - **Conflicting launch claims.** Gigazine says only SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium were included at the 2026-08-11 launch. Davies Meyer says Pro, Pro+ and the lower SuperGrok tiers were included "at launch". The x.ai news post presents those tiers as an expansion "today" (Aug 26, 2026), which matches Gigazine. - **Vellum's $200 standalone price and 14-day trial** do not appear in any official page in this corpus. Vellum also lists only Heavy / Ultra / Teams Premium access, which reflects the launch-era plan set. - **Grok Business is documented only on cursor.com.** The x.ai docs don't mention it, so verify on the canonical plans page. - **Legacy Privacy Mode blocks start-up.** Grok Bot requires cloud data storage, so the account setting must change first ([[concepts/privacy-and-data]]). - **@bot on X may stay quiet** when Grok Bot isn't linked to your Grok account or your weekly usage is used up ([[concepts/tag-bot-on-x]]). ## Related Concepts [[entities/grok-bot]], [[concepts/teams-and-enterprises]], [[concepts/team-bots]], [[concepts/settings-and-notifications]], [[concepts/mobile-app]], [[summaries/release-history]] ## Sources - raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md — https://x.ai/news/grok-bot-more-plans - raw/web_community-overview-md.md — https://docs.x.ai/grok-bot/overview.md - raw/web_community-grok-bot-md.md — https://cursor.com/docs/grok-bot.md - raw/web_community-get-started-md-2.md — https://docs.x.ai/grok-bot/get-started.md - raw/web_community-get-started-md.md — https://cursor.com/docs/grok-bot/get-started.md - raw/web_community-faq-md.md — https://docs.x.ai/grok-bot/faq.md - raw/web_community-teams-and-enterprises-md.md — https://docs.x.ai/grok-bot/teams-and-enterprises.md - raw/web_community-teams-md.md — https://cursor.com/docs/grok-bot/teams.md - raw/web_community-settings-and-notifications-md.md — https://docs.x.ai/grok-bot/settings-and-notifications.md - raw/web_community-team-bots-md.md — https://docs.x.ai/grok-bot/team-bots.md - raw/web_community-tag-on-x-md.md — https://docs.x.ai/grok-bot/tag-on-x.md - raw/web_community-troubleshooting-md.md — https://docs.x.ai/grok-bot/troubleshooting.md - raw/web_community-mobile-md.md — https://docs.x.ai/grok-bot/mobile.md - raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md — https://gigazine.net/gsc_news/en/20260812-spacexai-grok-bot/ (third-party) - raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md — https://ai-solutions.daviesmeyer.com/en/blog/grok-bot-ki-teammates-xai (third-party) - raw/web_community-official-grok-bot-breakdown-2026.md — https://www.vellum.ai/blog/official-grok-bot-breakdown (third-party) --- title: "Privacy, Data Handling, Training, and Retention" type: concept tags: [privacy, privacy-mode, training, retention, deletion, data-residency, models, zdr] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-approvals-security-and-privacy-md.md", "raw/web_community-security-md.md", "raw/web_community-security-md-2.md", "raw/web_community-security-faq-md.md", "raw/web_community-security-faq-md-2.md", "raw/web_community-faq-md.md", "raw/web_community-teams-md.md", "raw/web_community-teams-and-enterprises-md.md", "raw/web_community-computers-md.md", "raw/web_community-computers-md-2.md", "raw/web_community-inference-hooks-md.md", "raw/web_community-troubleshooting-md.md", "raw/web_community-work-md.md", "raw/web_community-official-grok-bot-breakdown-2026.md"] --- # Privacy, Data Handling, Training, and Retention ## Definition Grok Bot uses **Cursor authentication and Cursor account data settings**. Privacy, training, and retention follow the applicable Cursor account settings, privacy policy, and terms. Grok Bot **requires data storage** and does **not support Privacy Mode (Legacy)**, also called "Legacy Privacy Mode". ## How It Works **Training.** - Training opt-out follows the applicable Cursor account and privacy settings. The sources state no training behavior for the case where Privacy Mode is not enabled beyond that. - While a member is on a team, the team's privacy mode governs them. **With Privacy Mode enabled, customer data is not used for training.** Cursor enforces this on its servers, and "when the setting can't be verified, the system defaults to not training" (`teams.md`, both hosts). **Model providers.** - **Zero Data Retention** follows Cursor's existing provider agreements: providers don't keep prompts or outputs, and Grok Bot adds no separate control. - Providers *may* run abuse and safety classifiers, and flagged data may be stored for investigation. - Model selection is managed for you. There is no customer-facing model picker, and the serving mix can change over time. **On the computer.** - The durable disk keeps local files, browser sessions, and browser-saved data across sessions. - Connector tokens are never stored on the computer. - Idle computers hibernate, and hibernation is not deletion. - Cursor's endpoint telemetry on the computers "deliberately excludes customer data". **Retention and deletion.** - Under the DPA, data is deleted or returned **within 30 days of written direction after the service ends**. - The production control plane has **daily encrypted backups**, replicated to a separate recovery facility. - A per-organization retention policy and customer-managed point-in-time restore of a computer are **not available**. - Backend retention and account deletion follow the applicable Cursor terms. **Residency.** Computers run **in the United States today**. Cursor's US-only data residency program **does not apply to Grok Bot by default**. Ask your account team for a written commitment. ## Key Parameters | Item | Value | |---|---| | Privacy Mode (Legacy) | Not supported. It blocks Grok Bot entirely. Sign-in shows a Legacy Privacy Mode error | | Training with Privacy Mode on | Customer data not used | | Action Recording retention | **90 days** (Enterprise, **off by default**). Privacy Mode (Legacy) forces it off | | Conversation content export (xAI host) | Enterprise, **off by default**, opt-in. **Prompts** and **Responses** are exported. **Tool I/O** stays off until you enable it. Credentials, PEM blocks, and emails are redacted, but "pattern matching cannot catch every secret". Not available on Privacy Mode (Legacy) | | DPA deletion | Within 30 days of written direction after service ends | | **Delete All Grok Bot Data** (admin) | Permanent. At most 100 members per operation. The Cursor account stays | | Inference-hook frames | Cursor discards them after the verdict. Your endpoint's retention is yours (see [[concepts/inference-hooks]]) | ## When To Use Member clean-up when a project or login should no longer be available (docs.x.ai): 1. Pause or delete related routines. 2. Sign out of websites on the shared computer. 3. Uninstall connectors and revoke their authorization in the source service. 4. Remove sensitive project files from `/workspace`. 5. Hide or delete Bots that should no longer appear. 6. Use the account settings flow to delete the Cursor account if needed. ## Risks & Pitfalls - **Deleting a Bot doesn't remove shared-computer files or browser sessions.** - **Shared Bot templates** expose the configuration (identity, description, skills, routines). Strip API keys, internal URLs, and customer data first. Recipients never get your computer, logins, or history. - **Source conflict: the model allowlist.** - cursor.com: "The team model allowlist does not govern Grok Bot". Grok Bot may use third-party models from sub-processor-list providers regardless of it. The FAQ answer to "Can I restrict which models Grok Bot uses?" is **"No."** - docs.x.ai: the allowlist "is Enterprise only, and enforcement is not guaranteed… The list is honored by default", with an onboarding acknowledgement that it may not be followed. - Contact your account team if your contract restricts subprocessors. - **Source conflict: usage reporting.** - cursor.com: usage records the activity "as Grok Bot usage rather than by model". - docs.x.ai: "Usage analytics show the model that served each request, including failovers, and billing follows the serving model". - **Source conflict: delete-all scope.** xAI's list adds connected apps, email inboxes, files, and Team Bots that teammates use. Cursor's lists Bots, conversations, memory, routines, and computer. - **Third-party view (Vellum, medium confidence):** Vellum says xAI does not publicly document data-retention duration, memory export, or a memory API. It also says you "can't inspect, correct, export, or delete individual memories". That is Vellum's reading of the public docs. Official docs do say to correct stale assumptions by telling the Bot directly. ## Related Concepts - [[syntheses/security-model]], [[concepts/approvals]], [[concepts/inference-hooks]] - [[concepts/plans-and-access]], [[concepts/teams-and-enterprises]], [[concepts/cloud-computer]] ## Sources - raw/web_community-approvals-security-and-privacy-md.md (https://docs.x.ai/grok-bot/approvals-security-and-privacy.md) - raw/web_community-security-md.md (https://cursor.com/docs/grok-bot/security.md) - raw/web_community-security-md-2.md (https://docs.x.ai/grok-bot/security.md) - raw/web_community-security-faq-md.md (https://cursor.com/docs/grok-bot/security-faq.md) - raw/web_community-security-faq-md-2.md (https://docs.x.ai/grok-bot/security-faq.md) - raw/web_community-faq-md.md (https://docs.x.ai/grok-bot/faq.md) - raw/web_community-teams-md.md (https://cursor.com/docs/grok-bot/teams.md) - raw/web_community-teams-and-enterprises-md.md (https://docs.x.ai/grok-bot/teams-and-enterprises.md) - raw/web_community-computers-md.md (https://cursor.com/docs/grok-bot/computers.md) - raw/web_community-computers-md-2.md (https://docs.x.ai/grok-bot/computers.md) - raw/web_community-inference-hooks-md.md (https://cursor.com/docs/grok-bot/inference-hooks.md) - raw/web_community-troubleshooting-md.md (https://docs.x.ai/grok-bot/troubleshooting.md) - raw/web_community-work-md.md (https://cursor.com/docs/grok-bot/work.md) - raw/web_community-official-grok-bot-breakdown-2026.md (https://www.vellum.ai/blog/official-grok-bot-breakdown), third-party --- title: "Private Networks (Desktop Routing, Team Setup, Tailscale, Cloudflare Tunnel)" type: concept tags: [networking, private-network, team-setup, tailscale, cloudflare-tunnel, egress, network-policy] created: 2026-10-21 updated: 2026-10-21 confidence: high sources: ["raw/web_community-private-networks-md-2.md", "raw/web_community-private-networks-md.md", "raw/web_community-security-md.md"] --- # Private Networks (Desktop Routing, Team Setup, Tailscale, Cloudflare Tunnel) ## Definition How Bots reach internal systems (internal APIs, source control, databases, staging) from hosted computers that run in Cursor's cloud and egress through **shared static egress IPs**. There are two methods: route one member's traffic through their desktop, or install your own networking client on every team computer with **Team Setup** (Cursor's page calls it **Team Setup Scripts**; **Enterprise only**). Documented on `docs.x.ai/grok-bot/private-networks` and `cursor.com/docs/grok-bot/private-networks` with the same substance; the manifest UI path and naming differ (see below). ## How It Works | Method | Use it when | | --- | --- | | Route traffic through a desktop | One member needs access through a network already available from their device | | Install a networking client with Team Setup | Your Enterprise team needs a consistent connection on every hosted computer | **Desktop routing.** In the desktop app turn on **Route traffic through this computer**. xAI's page puts it at **Settings → Computer → Network**; Cursor's page says **Settings > Computer**, under **Your computer**. The route uses that device's network and IP and stops when the setting is off or an Enterprise admin disables **Allow Local Egress Routing** (see [[concepts/settings-and-notifications]]). **Team Setup** is a pattern *you* run, not a Cursor-managed network mode: Cursor runs your scripts; you install, configure and authenticate the client and maintain access rules. Cursor installs nothing by default and the dashboard shows no client status. It lives on the dashboard Grok Bot page (Cursor: under **Bot Computers**). Each manifest has entries with an **ID**, a **Setup Script**, and an optional **Check Script**. Group-scoped manifests go under a group's Grok Bot tab ([[concepts/teams-and-enterprises]]). Script execution rules: - Run as the computer user on every team computer; `sudo` is available for privileged installs. - Run at computer start and on a periodic refresh, roughly daily. - If the Check Script exits 0, the Setup Script is skipped; after setup runs, the check runs again to verify. - Entries run one at a time, in order; **30-minute timeout** per script; a failure doesn't block the computer and is retried later. - Scripts read **Team Secrets** as env vars; output is redacted; a computer with setup from more than one team gets no secrets. ## Key Parameters **Create a manifest:** dashboard Grok Bot page → **Team Setup** → next to **Manifests** select **+** (**New Manifest**) → **Manifest ID** (e.g. `private-network`) → entry **ID** (e.g. `install-network-client`) + **Setup Script** → optional **Check Script** (e.g. `command -v `) → **Save**. Toggle **Form** / **JSON** to edit as JSON: ```json { "manifestId": "private-network", "entries": [ { "id": "install-network-client", "setup": "#!/usr/bin/env bash\nset -euo pipefail\n# 1) Install your networking client per your provider's Linux install docs\n# 2) Start the client and connect it to your network per your configuration", "check": "command -v " } ] } ``` **Tailscale** (the configuration the pattern "has been exercised with"): run Tailscale on each computer and route via an exit node inside your network. Setup: install per Tailscale's Linux docs; Check: `command -v tailscale`. Authenticate via the login URL in the computer's browser, or store a reusable, tagged auth key as a Team Secret: `tailscale up --auth-key "$TS_AUTHKEY"` (all computers then join under the same key and tags, not member identity — scope ACLs). Subnet routers are possible but should be validated on a pilot first. **Cloudflare Tunnel** (not yet exercised the way Tailscale has; pilot first): a `cloudflared` connector inside your network publishes services through Cloudflare's edge behind Cloudflare Access. Setup: install `cloudflared`; Check: `command -v cloudflared`. HTTP services are reached by routed hostname; TCP services (databases, SSH) via `cloudflared access tcp`, which opens a local listener. Store Access service tokens as Team Secrets. **Choosing:** Tailscale gives network-level reach and, with an exit node, egress from your network. Cloudflare Tunnel is per-service and doesn't change the egress addresses others see. Other clients that run on Debian-based Linux follow the same steps. **Network policy interplay** ([[syntheses/security-model]]): Network Controls (Enterprise) is a separate layer — private reach doesn't replace your destination allowlist. In **Team Allowlist Only**, add the client's coordination servers, relays and gateways. Policy changes reach running computers within about a minute; sleeping computers apply them on wake; no recreate needed. Destinations can be web domains or IP ranges with ports. Static egress ranges are shared across Grok Bot customers, dedicated per-customer IPs are not available, and current ranges come from your account team (these docs publish no IP values). ## When To Use Use desktop routing for one-off access by a single member; use Team Setup for fleet-wide, consistent access on Enterprise. Prefer interactive browser login where your vendor supports it, since your IdP's policies apply ([[concepts/identity-and-access]]). **Rollout to existing computers:** new computers apply manifests at start; running ones on the ~daily refresh; to apply immediately, recreate the computer or have the member reset it from the desktop app. Image updates recreate computers and scripts re-apply. ## Risks & Pitfalls - **Never paste secrets into scripts** — manifests are plain text applied fleet-wide; use a Team Secret, e.g. `"$MY_AUTH_KEY"`. - **Client installed but nobody authenticated** (browser login is manual), or the auth key's Team Secret is missing, misnamed or expired. - **Allowlist-only policy blocks** Tailscale coordination/relays or Cloudflare endpoints. - Exit node not advertised/approved; Cloudflare connector down; Access denies the request. - **Recreated computers may lose sessions**, including the network client's login ("may need to be re-established") — re-authenticate if so. - `cloudflared access tcp` listeners don't persist across sessions. - **No fleet view of script results**; verify on a pilot computer. - **Not the Cloud Agents setup:** Cloud Agent recipes (userspace networking, proxy variables) are a different surface; delegated Cloud Agents use Cloud Agent network settings, not your computer's client. ## Related Concepts - [[concepts/proxies]] (device-to-Cursor path, the opposite direction), [[concepts/cloud-computer]], [[concepts/teams-and-enterprises]], [[syntheses/enterprise-rollout-guide]] ## Sources - raw/web_community-private-networks-md-2.md — https://docs.x.ai/grok-bot/private-networks.md - raw/web_community-private-networks-md.md — https://cursor.com/docs/grok-bot/private-networks.md - raw/web_community-security-md.md — https://cursor.com/docs/grok-bot/security.md (Network policy, Static egress IPs sections) --- title: "TLS-Inspecting Proxies (Zscaler, Allowlists, cursorvm.com)" type: concept tags: [proxies, zscaler, tls-inspection, allowlist, cursorvm, networking, troubleshooting] created: 2026-10-21 updated: 2026-10-21 confidence: high sources: ["raw/web_community-proxies-md-2.md", "raw/web_community-proxies-md.md"] --- # TLS-Inspecting Proxies (Zscaler, Allowlists, cursorvm.com) ## Definition Gateway configuration for the path **from a member's device to Cursor**. The desktop app connects to two places: Cursor's API at `*.cursor.sh` (chat, sign-in, approvals) and the member's hosted computer at a nested `*.*.cursorvm.com` hostname (computer setup, screen, shell). Secure web gateways that inspect TLS often let the first through and break the second. Zscaler is the worked example, but the steps apply to any gateway that re-signs TLS or buffers responses. The xAI (`docs.x.ai/grok-bot/proxies`) and Cursor (`cursor.com/docs/grok-bot/proxies`) pages are effectively identical. This page does **not** control where the hosted computer itself may connect — that is the network policy — or reaching private services from the computer ([[concepts/private-networks]]). ## How It Works | Pattern | Used for | | --- | --- | | `*.cursor.sh` | Chat, sign-in, approvals, and the rest of the Cursor API | | `*.cursor-cdn.com` | Static assets | | `*.cursorapi.com` | Extension marketplace and related APIs | | `*.cursorvm.com` | The hosted computer and its control plane | | `*.*.cursorvm.com` | The same, one level deeper. Required. | | `cursor.com`, `downloads.cursor.com` | Installing and updating the desktop app | Allow all of these on the gateway **and** in DNS filtering, using wildcards (computer hostnames are generated per computer). Computer hostnames have the form `..cursorvm.com`, so a single-level `*.cursorvm.com` alone looks right but leaves the computer unreachable. On every domain above: - **Bypass TLS (SSL) inspection** — a re-signed certificate breaks the computer setup handshake even when the host is allowed. - **Turn off response buffering** — chat and the computer link stream. If policy demands inspecting everything, the gateway must support HTTP/2 or Cursor's HTTP/1.1 fallback, Server-Sent Events passthrough without buffering, and long-lived connections without forced timeouts. **Apply to every profile, including off-network.** Zscaler Client Connector applies its own off-network profile; exceptions added only to the office location don't follow the device home. Apply allow rules, inspection exemption and DNS exceptions to every location, profile and policy group, including roaming. ## Key Parameters Verify from a member's device with the gateway client running: ```bash curl -v https://api2.cursor.sh |& grep -C1 issuer: ``` Issuer should be **Amazon RSA**; if it names Zscaler or your vendor, inspection is still active on `cursor.sh`. ```bash nslookup test.us9.cursorvm.com ``` Should return addresses. If it fails, try `nslookup test.us9.cursorvm.com 1.1.1.1`; if the public resolver answers and yours doesn't, the nested `*.*.cursorvm.com` exception is missing from the device's DNS or gateway profile. Then run the HTTP/1.1 and HTTP/2 streaming tests from Cursor's Enterprise network configuration page (output should arrive line by line), and retry connecting in the app. **Scope table:** | You want to | Use | | --- | --- | | Let devices reach Cursor and the hosted computer through your gateway | This page | | Limit which destinations the hosted computer may reach | Network policy, Enterprise only ([[syntheses/security-model]]) | | Route hosted-computer traffic through a member device | Route traffic through your desktop ([[concepts/settings-and-notifications]]) | | Install a networking client on every hosted computer | Team Setup, Enterprise only ([[concepts/private-networks]]) | By default the hosted computer's own traffic leaves from Cursor's shared static egress IPs and does not pass through the member's gateway. When a member enables desktop routing, routed traffic uses that device's network and is subject to its gateway policy. (Naming differs: xAI calls the toggle **Route egress through this desktop** here, Cursor **Route traffic through this computer**.) ## When To Use Before pushing the desktop app to any fleet behind a secure web gateway — step 2 of the rollout in [[concepts/enterprise-deployment]] and [[syntheses/enterprise-rollout-guide]]. The list is the same as Cursor's Enterprise network configuration list; the editor works without the `cursorvm.com` patterns, which is why a gateway fine for the Cursor editor can still break Grok Bot. ## Risks & Pitfalls Symptoms and causes: - **Computer setup hangs or fails; chat may keep working** (it reaches `api2.cursor.sh`) — the computer link needs `cursorvm.com`. - **Works on a hotspot/personal device, fails on corporate network** — gateway is the cause. - **Works in office, fails at home** — exceptions applied to the office location only. - **Sign-in or chat stall too** — inspection or buffering still active on `cursor.sh`. - **"We allowlisted `*.cursorvm.com`"** — the nested `*.*.cursorvm.com` is missing, or the domain is allowed but still TLS-inspected. This is the usual failure on networks where `cursor.sh` is already allowed. For broader symptom triage see [[syntheses/troubleshooting]]. ## Related Concepts - [[concepts/enterprise-deployment]], [[concepts/private-networks]], [[concepts/teams-and-enterprises]], [[concepts/cloud-computer]] ## Sources - raw/web_community-proxies-md-2.md — https://docs.x.ai/grok-bot/proxies.md - raw/web_community-proxies-md.md — https://cursor.com/docs/grok-bot/proxies.md --- title: "Settings and Notifications" type: concept tags: [grok-bot, settings, notifications, auto-review, updates, timezone] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-settings-and-notifications-md.md", "raw/web_community-settings-md.md", "raw/web_community-mobile-md.md", "raw/web_community-team-bots-md.md", "raw/web_community-work-md.md"] --- # Settings and Notifications ## Definition The **Grok Bot settings** dialog holds app-wide and desktop-local behavior. Per-Bot profile and notification options live in each Bot's conversation details. Open settings from the account menu (**Settings**) or with `Cmd/Ctrl+,` (cursor.com writes Cmd+,). cursor.com names the panes **General**, **Computer**, **Usage & Billing**, and **Updates**. Both hosts warn that sections depend on your account and rollout, so some options may not appear. ## How It Works ### General - **Account:** sign in or out of the Cursor account Grok Bot uses. The account menu also offers **Get Grok Bot for mobile** and shows the version. On docs.x.ai that is **About**; on cursor.com it is **Support** > **About**. With more than one saved account, the menu offers **Switch account** and **Add account**, and inactive accounts have **Remove** (docs.x.ai). - **Appearance:** **Follow System**, **Light**, or **Dark**. docs.x.ai also lists **Language**: **Follow System** or more than 20 languages. On mobile the option is labeled **System** and the language list is shorter. - **Bot:** **Timezone** (used for routine schedules). There is no model picker, because model selection is managed for you. - **Auto-review:** your personal rules (cursor.com: an Auto-review switch plus **Custom Rules**) and any team rules your admin requires. Locked team rules show `Required by your admin. You can't edit or delete this rule.` Your own rules can only make behavior **stricter**, and **Ask first** wins when rules conflict. If the admin turns enforcement off, you see only your own rules. - **Team Bots** (under **Settings → General → Team Bots**, per `team-bots.md`): link your Slack account, and **Clear** remembered connector approvals. ### Computer - **Execution on Local Computer** controls whether Bots can run commands on the desktop in front of you. **Per-command approval is the default**, and the setting applies to that desktop only. cursor.com shows it as **Execution on this computer** per listed desktop under **Your computer**, with options always allow, ask every time, or never (`work.md`). docs.x.ai places it under **General → Bot**. - **Route traffic through this computer** sends the cloud computer's web traffic through this desktop, so destinations see your desktop's IP. It applies to one desktop and, per cursor.com, to new connections. If an Enterprise admin turns off **Allow Local Egress Routing**, the toggle locks with `Disabled by your admin. You can't turn this on.`, and any active route stops within five minutes. Your choice is kept for when the admin re-allows it (see [[concepts/private-networks]]). - **Bot's computer:** **Update Bot's Computer** and **Reset Bot's Computer**. ### Usage & Billing, Updates, Plugins - **Usage & Billing** shows weekly included usage and on-demand usage for eligible accounts. The account menu can show **Weekly usage** (see [[concepts/plans-and-access]]). - **Updates:** **Check for Updates** and **Restart to Update** update the desktop app. The app and the cloud computer update separately. - **Plugins/apps are not a settings section.** Use **Connect apps** in the sidebar (docs.x.ai: **Your plugins** → **Manage plugins and skills** lists **Installed** and **Private skills**). - **Team Setup Scripts** are Enterprise only. Admins manage them from the dashboard, and members don't see them in settings (cursor.com). ### Per-Bot settings Open **View conversation details** → **Bot settings** (`Cmd/Ctrl+Shift+,` is the Bot settings shortcut) to edit **Name**, **Label (optional)**, avatar, and **Notifications**. docs.x.ai also lists **Description**, but cursor.com says personal Bots have no description (see [[concepts/bots]]). ### Attention states The sidebar shows **Needs attention** (a question, approval, or handoff), **Unread activity** (a new result), and working or typing status. Opening a conversation marks it read, and the Bot menu can mark a conversation read or unread manually. ### Notifications Turn on **Notifications** for a Bot to get an OS or mobile notification when it finishes or needs input. Group chats have no per-Bot switch. Where the switch lives differs by host: docs.x.ai puts it in the Bot's settings, while cursor.com says it is in the Bot's profile on the phone and that on desktop you **ask the Bot** to turn notifications on or off. Routing (docs.x.ai): while you're using the desktop app on macOS or Windows, notifications from Bots you own go to the desktop instead of your phone. "In use" means the app is open and signed in, the screen is unlocked, you've used the computer in the last 5 minutes, and system settings allow Grok Bot banners or alerts. A focused window silences only the Bot on screen. Finished replies skip your phone. When a Bot needs input, your phone is notified about 3 minutes later unless you open that Bot, turn off its Notifications, or hide it first. At other times, and always on Linux, notifications are suppressed while Grok Bot is focused and the phone is notified as usual. cursor.com gives only the simpler rule (suppressed while focused) and adds that iPhone push delivery is rolling out gradually. On mobile, both the device permission and the Bot's setting must allow a notification. ### In-app errors Errors appear above the composer under **Notifications**. You can dismiss one notice or clear the list. **Copy request ID** gives support the complete ID. Clearing a notice does not undo the underlying action or history. ## Key Parameters | Setting / value | Default or limit | |---|---| | Open settings | `Cmd/Ctrl+,` | | Execution on Local Computer | per-command approval (default), per desktop | | Local-egress admin lock takes effect | within five minutes | | Desktop "in use" window | used in the last 5 minutes | | Phone fallback for needs-input | about 3 minutes | | App languages | more than 20 (desktop) | | Auto-review conflict rule | **Ask first** wins | ## When To Use - Set **Timezone** before creating routines. - Use **Route traffic through this computer** when a Bot must reach sites or networks only your desktop can reach. - Turn on per-Bot **Notifications** for Bots whose results or approvals are time-sensitive. - Read [[concepts/approvals]] before loosening local execution or editing Auto-review rules. ## Risks & Pitfalls - **Where Auto-review rules are stored is disputed.** docs.x.ai says personal rules are stored on the current desktop and synced to its Grok Bot computer, so another desktop needs its own. cursor.com says they are saved to your account and apply on every desktop and on iPhone. Verify in-app. - **Reset Bot's Computer is a last resort.** It rebuilds the computer from your last saved snapshot or data, so very recent changes may be lost. A *computer* update (as opposed to a software update) removes installed apps and packages but keeps files and logins (cursor.com). Follow the least-destructive order in [[syntheses/troubleshooting]]. - Local routing exposes your desktop's IP and networks to the Bot. - Notifications won't arrive if device permission is off, even when the Bot setting is on. ## Related Concepts - [[concepts/bots]] - [[concepts/approvals]] - [[concepts/cloud-computer]] - [[concepts/mobile-app]] - [[concepts/skills-routines-and-automations]] - [[concepts/plans-and-access]] ## Sources - `raw/web_community-settings-and-notifications-md.md`: https://docs.x.ai/grok-bot/settings-and-notifications.md - `raw/web_community-settings-md.md`: https://cursor.com/docs/grok-bot/settings.md - `raw/web_community-mobile-md.md`: https://docs.x.ai/grok-bot/mobile.md - `raw/web_community-team-bots-md.md`: https://docs.x.ai/grok-bot/team-bots.md - `raw/web_community-work-md.md`: https://cursor.com/docs/grok-bot/work.md --- title: "Skills, Routines, and Automations" type: concept tags: [grok-bot, skills, routines, schedules, triggers, teach-a-task] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-skills-routines-and-automations-md.md", "raw/web_community-work-md.md", "raw/web_community-mobile-md.md", "raw/web_community-team-bots-md.md", "raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md"] --- # Skills, Routines, and Automations ## Definition Grok Bot turns successful one-off work into repeatable processes using two building blocks: - A **skill** is a reusable set of instructions for *how* to do a task: steps, decision rules, expected output, and safety boundaries. - A **routine** tells **one Bot** *when* to run a workflow: on a schedule or, where supported, after an event. The recommended progression on both hosts: do the task once, make it reliable, save the method as a skill, and only then automate it with a routine. ## How It Works ### Save a skill Ask the Bot directly: > Save the process we just used as a skill called “Weekly account health.” Include the source systems, risk definitions, output format, and the rule that customer contact always requires approval. A useful skill states: (1) when to use it, (2) required inputs and access, (3) the sequence of work, (4) how to validate the result, (5) what to return, (6) what requires approval. Skills are available across your Bots, although a Bot may still need the relevant connector or login. Reference a skill with `/` in the composer. The `@` menu covers Bots, groups, routines, and connectors. **Connect apps** in the sidebar is where you discover supported connectors and packaged skills. **If a private skill is missing from the `/` menu**, the two hosts give different paths: - docs.x.ai: **Connect apps → Your plugins → Manage plugins and skills**, then check it is listed under **Private skills**. Private skills are described as "one library shared by all your Bots." - cursor.com: **Connect apps**, select the connected count next to the title, and **enable it for the current Bot** under **Private skills**. Wiki inference: this suggests cursor.com enables skills per Bot, while docs.x.ai describes one private-skill library shared by all your Bots. ### Teach a task (demonstration) 1. Open a one-to-one Bot conversation and its computer view. 2. Choose **Teach a task** and describe the result you're about to demonstrate. 3. Perform the browser workflow once. 4. Stop recording and review the **draft skill** the Bot creates. 5. Add decision rules, failure handling, and approval boundaries, then test on a safe example before scheduling. Recording captures visible computer interaction for **up to ten minutes** and does **not** record microphone audio. Teach a task is rolling out gradually; if the control isn't visible, ask the Bot to create a skill from written instructions and a completed task. It isn't available on mobile. Note: the Davies Meyer blog (third-party) says a demonstrated process is stored as a *routine*. The official docs say demonstration produces a *skill*, and routines are a separate scheduling layer. ### Create a routine Ask the Bot that should own the job: > Every weekday at 8:00 AM, run the Daily customer-risk skill against the current account list. Post a linked watch list in this conversation. Do not contact customers. If the source data is unavailable, report the failure instead of using old data. Confirm the owning Bot, the schedule and **time zone**, the input source, the expected result, the approval boundary, and what happens when a source is missing. The Bot creates the routine and shows its next run. Routines run in the cloud while your laptop is closed. The time zone comes from **Settings → General → Bot → Timezone** (see [[concepts/settings-and-notifications]]). ### Event triggers **Cursor account integrations** can start a routine from an event such as a Slack message or a GitHub notification. They are separate from the Slack or GitHub plugins/apps and may need their own connection flow. Use a narrow matching rule, for example: when a message in `#customer-escalations` contains a support ticket link and the phrase “needs repro”, reproduce the issue in staging and post a repro pack, and never post back to Slack without approval. cursor.com also points to schedules, Slack triggers, and webhooks on its Routines help page, which is not in this corpus. ### Test and manage - docs.x.ai: use **Test run** after creating or editing. Manage via **View conversation details → Routines**: enable or pause, run a test, edit schedule or instructions, inspect recent success and failure history, delete. - cursor.com: create and change routines by asking the Bot in chat. On desktop, click the Bot's name to open its details; routines are on the **Details** tab. Each routine shows **Instruction** and **When to run**, with **Pause**, **Resume**, **Test**, **Edit**, and delete. - Mobile: the Bot's profile lists routines with schedule, next run, instruction, and **Run history**. **Active** pauses or resumes a routine, and you can delete from there. **Editing and testing require the desktop app.** When reviewing a test, check that it used current inputs, met the output format, left a source or audit trail, stopped at the intended approval point, and made failure states explicit. ## Key Parameters | Parameter | Value (source) | |---|---| | Max routines per Bot | 50 (docs.x.ai) | | Run records kept per routine | 20 most recent (docs.x.ai) | | Minimum schedule spacing | at least five minutes apart (docs.x.ai) | | Teach-a-task max recording | ten minutes, no microphone audio | | Skill reference | `/` | | Mention routine/connector | `@` | | Delete routine | immediate, no undo | | Team Bot routines | personal: each runs as, reports to, and is visible only to the person who set it up | ## When To Use - **Skill:** you've done a task successfully and want it done the same way again, by any Bot. - **Routine (schedule):** reliable recurring preparation work, such as digests, watch lists, and reconciliations. - **Routine (event):** a narrowly defined trigger with a clear response. - **Team Bot output for everyone:** have a personal routine post its result somewhere shared, such as a Slack channel (see [[concepts/team-bots]]). ## Risks & Pitfalls - **A test run performs real work.** It can navigate websites, change files, and call connected tools. Use safe inputs and keep write actions behind approval. - **Broad listeners** ("every new message") create noise, consume usage, and act on irrelevant input. - **Deleting a routine can't be undone.** Deleting a Bot also removes the routines it owns. Hiding a Bot does **not** pause its routines. - **Auto-pause:** after a long period away, Grok Bot may ask whether to keep routines running and pause them if you don't respond. Review paused routines when you return. - **Teach-a-task drafts are incomplete.** One demonstration doesn't capture edge cases. Keep secrets out of the recording and use the secure secret request for credentials. - **Drift:** re-test after a website, connector, or source format changes. - Design for trust: automate preparation before execution, require approval for sending, purchasing, deleting, publishing, or production changes, include a no-data and stale-data policy, make retries idempotent, and say where partial completion should be reported (see [[concepts/approvals]]). ## Related Concepts - [[concepts/bots]] - [[concepts/team-bots]] - [[concepts/apps-and-connectors]] - [[concepts/settings-and-notifications]] - [[concepts/use-cases]] - [[concepts/mobile-app]] ## Sources - `raw/web_community-skills-routines-and-automations-md.md`: https://docs.x.ai/grok-bot/skills-routines-and-automations.md - `raw/web_community-work-md.md`: https://cursor.com/docs/grok-bot/work.md - `raw/web_community-mobile-md.md`: https://docs.x.ai/grok-bot/mobile.md - `raw/web_community-team-bots-md.md`: https://docs.x.ai/grok-bot/team-bots.md - `raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md`: https://ai-solutions.daviesmeyer.com/en/blog/grok-bot-ki-teammates-xai (third-party) --- title: "Tag @bot on X" type: concept tags: [grok-bot, x, twitter, at-bot, regional-availability] created: 2026-10-21 updated: 2026-10-21 confidence: high sources: ["raw/web_community-tag-on-x-md.md", "raw/web_community-overview-md.md", "raw/web_community-troubleshooting-md.md"] --- # Tag @bot on X ## Definition When you tag **@bot** (https://x.com/bot) in a post or reply on X, that post is handed to your Grok Bot as a task. @bot answers on X with a short reply confirming that your Grok Bot has the request, and your **main Bot** starts working on its cloud computer. The work and its results stay in that Bot's conversation in the Grok Bot app. **@bot posts only the confirmation on X.** This feature is documented only on docs.x.ai (`tag-on-x.md`, with a summary in `overview.md`). No cursor.com page in the corpus covers it. ## How It Works Example reply on a thread: > @bot summarize this thread and send the key points to my Slack **What the Bot receives:** your post, the post you replied to, and any post either of them quotes, including still photos. **First use:** the first time a request goes through, X sends you a notification that tagged requests start work automatically and use your Grok Bot usage. ### Setup (prerequisites) 1. A Grok account with your X account connected. On grok.com open **Settings → Account** and choose **Connect** next to **𝕏 Account**. 2. Grok Bot access on that **same** Grok account. Download the app from x.ai/bot. If your access comes from a SuperGrok subscription, link it as described in Get started (see [[syntheses/getting-started]] and [[concepts/plans-and-access]]). 3. At least one **Bot of your own**. Bots someone shared with you can't take tagged requests. 4. An X account based in a region where tagging is available. The **X for Grok Bot** plugin is optional. It lets your Bot search posts, read timelines, check mentions, and use trends and bookmarks. @bot forwards requests whether or not it is installed. ### Reading @bot's reply | @bot replies | Meaning | Action | |---|---|---| | A short line saying it sent your request to your Grok Bot | Your main Bot has the task | Open Grok Bot to follow the work and review the result | | "Sign up for Grok Bot: https://x.ai/bot" | No Grok account is connected to the X account you posted from | Connect X on grok.com, set up Grok Bot, tag again | | "Grok Bot isn't set up yet. Open https://x.ai/bot" | Accounts are connected but you have no Bot of your own | Create a Bot in Grok Bot, tag again | ## Key Parameters - **Region exclusions (as of fetch 2026-10-21):** not yet available for X accounts based in **Australia, the United Kingdom, the European Union, Iceland, Liechtenstein, or Norway**. - **How region is decided:** by the country X has on record for your account. Current location and VPNs have no effect. If X lists only a broader region (for example "South Asia") or no country, tagging isn't available either. - **Which Bot:** your main Bot. Shared Bots are not eligible. - **Usage:** tagged requests use your Grok Bot usage. If your weekly usage is used up, @bot stays quiet (listed among the causes of no reply). ## When To Use Use it to turn something you're reading on X (a thread, an announcement, a quoted post with photos) into a task without opening the app, for example summarizing a thread and sending it to Slack. Follow and review the work in the app, because the results never post back to X. Install the X for Grok Bot plugin if the task needs deeper X access, such as timelines, mentions, trends, or bookmarks. ## Risks & Pitfalls **When @bot stays quiet, your Bot did not get the request.** Common causes: - Your X account is in a region where tagging isn't available yet (including the region and no-country cases above). - You replied in a thread that already includes @bot but didn't tag @bot yourself. - The post also tags **@grok**. - The post, the post it replies to, or a quoted post contains a **video or GIF** (still photos are fine). - The post doesn't pass internal moderation. - Grok Bot isn't linked to your Grok account, or your weekly usage is used up. Other pitfalls: - **Work starts automatically** once a request is accepted, and it consumes usage. Wiki analysis: phrase the tag as you would a direct instruction, and keep external actions behind approval (see [[concepts/approvals]]). - The @bot confirmation is a short **public** reply on X (docs.x.ai overview.md); the work and results stay in the app. ## Related Concepts - [[concepts/bots]]: your main Bot receives the task - [[concepts/plans-and-access]]: SuperGrok linking and weekly usage - [[concepts/apps-and-connectors]]: the optional X for Grok Bot plugin - [[syntheses/troubleshooting]] - [[entities/grok-bot]] ## Sources - `raw/web_community-tag-on-x-md.md`: https://docs.x.ai/grok-bot/tag-on-x.md - `raw/web_community-overview-md.md`: https://docs.x.ai/grok-bot/overview.md - `raw/web_community-troubleshooting-md.md`: https://docs.x.ai/grok-bot/troubleshooting.md --- title: "Team Bots" type: concept tags: [grok-bot, team-bots, sharing, permissions, slack, managers, memory] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-team-bots-md.md", "raw/web_community-teams-md.md", "raw/web_community-work-md.md"] --- # Team Bots ## Definition A **Team Bot** is one Bot that the whole team talks to. Its owner sets it up once with plugins, secrets, skills, and files, then publishes it to the team. Every teammate can then chat with it in the Grok Bot app (desktop or mobile) or in Slack. Each teammate gets their own **private** chat: the owner can't read teammates' chats, and teammates can't read the owner's. cursor.com `teams.md` describes Team Bots as "the exception to one Bot per user". A Team Bot can be shared with everyone on the team or only with certain people and groups. Team Bots are available to members of a team. If an account can't use them, a Team Bot link shows **Team Bots Not Available**. ## How It Works | | Personal Bot | Public template | Team Bot | |---|---|---|---| | Who talks to it | Only you | Each adopter talks to their own copy | Everyone on the team, each in their own chat | | Who maintains it | You | Each copy's owner, separately | Owner and managers, and changes reach everyone | | Memory | Yours | Starts from the template, then separate per copy | Team memory plus private notes with each person | ### Create and publish 1. Choose **New chat** or press `Cmd/Ctrl+N`, then choose **Create new Team Bot**. The Bot asks what the team needs and walks you through plugins, secrets, skills, and files. Until it is published, only you can see it. 2. Or start from an existing Bot: **Share** menu → **Publish to Team**. If most of the Bot's memories are personal facts about you, it asks **How should your Team Bot start?** and offers **Copy** *Bot name* or **Start fresh**. Otherwise it replies **Done. Your team copy is in the sidebar.** Setup cards then let you choose which plugins, secrets, memories, skills, files, and routines the copy gets. Memories you don't pick stay private, and **Keep private** shares none. 3. Publish with **Publish to team** or by asking the Bot. Teammates find it under **New chat → Team Bots**, and **Copy link** shares it directly. **Unpublish** removes teammates' access. Their chats and routines come back when you publish again. ### Setup, credentials, and memory The **Setup** section has cards for **Plugins**, **Secrets**, **Skills**, and **Files**. Only the owner and managers can change it, and every change reaches all teammates. Whose account a plugin uses depends on how the plugin signs in: | Plugin type | Whose access | Teammate setup | |---|---|---| | OAuth sign-in | The person talking to the Bot | Each connects their own account the first time | | Key or token | The Bot's own credential | None | | Custom MCP, **Remote HTTPS** | The Bot's credential, or each person's OAuth sign-in | None, or a one-time sign-in | | Custom MCP, **Command** | Runs on the computer that conversation uses | None. If the command needs secrets, it works only in the owner's chat | - **Memory:** **Team memory** is read by every teammate's conversation. The Bot saves a fact there only when someone says the whole team should have it, and it tells you when it does. **Notes with each person** are private to that person and follow them across app chats and Slack DMs. - **Skills** apply in every teammate's chat. Only the owner or a manager saves a team skill. ### Where the work runs | Conversation | Computer | |---|---| | Owner's own chat | Owner's computer | | A teammate's app chat or 1:1 Slack DM | That teammate's computer | | Slack channels, group DMs, threads | One shared computer for that Bot | `teams.md` also lists group chats as using the Bot's own computer. `teams.md` also says that in a member's own chat the Bot uses that member's connected accounts "after asking", whereas docs.x.ai `team-bots.md` says the Bot uses a person's own connectors in their 1:1 chat "without asking" (a source divergence). ## Key Parameters | Parameter | Value | |---|---| | Max secrets per Team Bot | 25 | | Secret value length | at least 8 characters, at most 4,096 bytes | | Secret name format | uppercase letters, digits, underscores; starts with a letter (e.g. `DATADOG_API_KEY`) | | Secret redaction | values replaced with `[REDACTED]` in command output, files read, and plugin calls | | Team file formats | .txt, .md, .markdown, .csv, .json, .yaml, .yml; up to 256,000 characters per file | | Description | up to 140 characters, written in the Bot's own voice | | Search Team Bots (desktop) | type at least three characters | | Remembered connector approvals | cleared at **Settings → General → Team Bots** → **Clear** | | Slack account link | **Settings → General → Team Bots** | **Who can do what** | Action | Owner | Manager | Teammate | Team admin | |---|---|---|---|---| | Chat, pin, hide, create own routines | yes | yes | yes | — | | Change setup (profile, skills, plugins, secrets, Slack app) | yes | yes | no | no | | Publish / unpublish / delete | yes | no | no | delete only, via **Manage Team Bots** | | Add to Slack | yes | yes (uses owner's Slack connection) | no | — | Managers are added in the Bot's details under **Managers** → **Add** → **Add manager**, and removed with **Remove manager**. **Personal connectors in shared conversations.** In a Slack thread or group chat, the Bot asks before it uses your own connector: **Allow once**, **Always allow for this Bot**, **Always allow for all Team Bots**, or **Skip**. Your 1:1 chat and the Bot's own plugins don't trigger this prompt. **Usage.** A teammate's messages count against that teammate's own allowance, and a routine counts against whoever set it up. Slack messages that don't come from a linked teammate, such as workflow posts, count against the owner. If a teammate is out of usage, the Bot tells them instead of using the owner's allowance. **Admin controls** (Cursor dashboard → **Team Bots** → **Manage Team Bots**): - docs.x.ai `team-bots.md`: for each Bot, choose who gets it in their sidebar: **All team**, **None**, or specific groups. Those members see **Required by your admin. You can't turn this off.** - cursor.com `teams.md`: an **Access** column sets who can use the Bot (whole team, certain people or groups, or only its owner), and an **Auto-add** column (**All team** or groups) works only for groups that have access. Auto-added members can't hide the Bot. ## When To Use Use a Team Bot for one shared job: metrics questions from a data warehouse, keeping docs current as PRs ship, triaging a Slack channel, or answering helpdesk questions. Keep a personal Bot for an everyday assistant that works from your own accounts and memories. If most of what a Bot knows is about you personally, create a fresh Team Bot instead of copying it. ## Risks & Pitfalls - **Secrets are team-wide access.** The Bot can use every secret in any teammate's conversation. Use scoped, read-only service-account keys. - **Command MCP servers:** teammates can use the shell and read the command and arguments, so never put a credential in either. Use **Remote HTTPS** for servers that need secrets. - **Slack channel threads share one computer.** Don't share anything there that shouldn't be visible to everyone in the channel. - **Approvals with nobody to answer:** where nobody can answer an approval card, the Bot runs **without Auto-review** within its configured permissions, unless the team enforces Auto-review (see [[concepts/approvals]]). - **Routines are personal.** No routine runs for the whole team. To share a routine's output, have it post somewhere shared, such as a Slack channel. - **Delete can't be undone.** Deleting a Team Bot removes it for everyone, along with their chats and routines. Use **Unpublish** if you may need it again. - **Slack requires linking.** The Bot answers only people who have linked their Slack account to Cursor; workflow and app posts are the exception. A linked account that isn't on the Bot's Cursor team gets a private notice and no answer. In channels the Bot ignores top-level messages that don't @mention it; use a channel listener routine to have it respond to every message. Workspaces that require app approval show **Awaiting admin approval**. ## Related Concepts - [[concepts/bots]]: personal Bots and templates - [[concepts/teams-and-enterprises]]: admin controls - [[concepts/apps-and-connectors]]: plugin connection - [[concepts/skills-routines-and-automations]] - [[concepts/approvals]] - [[concepts/messaging-and-collaboration]]: Slack-thread handoffs ## Sources - `raw/web_community-team-bots-md.md`: https://docs.x.ai/grok-bot/team-bots.md - `raw/web_community-teams-md.md`: https://cursor.com/docs/grok-bot/teams.md - `raw/web_community-work-md.md`: https://cursor.com/docs/grok-bot/work.md --- title: "Teams and Enterprises (Admin Controls)" type: concept tags: [admin, teams, enterprise, dashboard, admin-controls, group-settings, team-secrets, audit] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-teams-and-enterprises-md.md", "raw/web_community-teams-md.md"] --- # Teams and Enterprises (Admin Controls) ## Definition How Grok Bot is made available to a Cursor team and which controls admins get. Two official pages cover this: xAI's `docs.x.ai/grok-bot/teams-and-enterprises` and Cursor's `cursor.com/docs/grok-bot/teams`. They are near-identical; the Cursor page is the more complete one (see divergences below). Almost every admin control lives on the **Grok Bot page** of the Cursor dashboard (`https://cursor.com/dashboard/bot`), which members never see. ## How It Works **Availability** (both pages): | Plan | Access | | --- | --- | | Individuals | Every paid Cursor plan, or an individual SuperGrok link. Cursor's page adds: "or with a seat on a self-serve Grok Business plan" | | Teams | Included; every member has access; usage follows the seat's allowance | | Enterprise | An admin enables it from the dashboard | Plan and usage details belong to [[concepts/plans-and-access]]; both pages call `https://cursor.com/help/grok-bot/plans` the canonical matrix. **Enabling.** On **Teams**, Grok Bot is on by default for every member and there is no switch to turn it off. It stays off for teams on **Privacy Mode (Legacy)** or a legacy request-based plan. On **Enterprise**, an admin turns it on (the organization-wide **Enable Grok Bot** switch) and scopes it to everyone or to specific groups with **Manage Group Access**. Turning it off blocks every member *without deleting their computers*. The setup modal covers privacy mode, pricing/billing and model availability; Cursor's page adds that it also offers to turn on Cloud Agents if they're off and asks you to accept the Grok Bot Terms. **Inviting.** Admins on either plan use **Invite Team** on the Grok Bot page. Existing Cursor users get an email with a download link; new users get a team invitation by email or invite link. Teams managed through SCIM only see the existing-users option. **Architecture an admin should know.** Each user gets one persistent Firecracker microVM; all of that user's Bots share it (Bots isolate personalities and workspaces, not compute). Logins and files on the computer are reachable by every Bot that user runs, so a workload needing its own credential set needs its own Cursor user. Cursor's page adds the exception: [[concepts/team-bots]] are shared with teammates and use a computer of their own in Slack channels and group chats. See [[concepts/cloud-computer]]. ## Key Parameters Admin controls ("Ent" = Enterprise only; others are Teams and Enterprise): | Control | Default / values | Plan | | --- | --- | --- | | Enable Grok Bot + **Manage Group Access** | org-wide switch | Ent | | SCIM | SCIM 2.0 via your IdP — see [[concepts/identity-and-access]] | Ent | | Cloud Agents (delegation) | on by default, team-wide | All | | Agent Email | off by default; turning it off removes email tools and stops email routines. Cursor: address is `mail.grokbot.com`, outbound mail waits as a draft until the member presses Send, existing addresses keep receiving mail | All | | Public template sharing | Enterprise starts off; other teams start allowed; enforced server-side | All | | Connector policy | inherited from Cursor Team Marketplace (`https://cursor.com/dashboard/plugins`); MCP allowlist is Ent; blocked connector shows **Disabled by team admin**; pushing connectors to members is not available | All | | Execution on Local Computer | **Always allow** (default, defers to member, whose own default is ask), **Ask every time**, **Never allow** | All | | Allow Local Egress Routing | on by default; turning off stops active routes within five minutes | Ent | | Allow Sudo on Bot Computers | on by default; existing computers must restart to pick up a change | Ent | | Team Rules | scoped to Cursor, Grok Bot, or both; Grok Bot rules are always required | All | | Enforce Auto-review | off by default | Ent | | Auto-review rules | **Ask first** / **Allow automatically** via **Configure Rules**; **Ask first** wins on conflict | Ent | | Inference Hooks (Cursor page only) | beta, new configs start in Shadow mode — see [[concepts/inference-hooks]] | Ent | | Network Controls (**Grok Bot Network Access**) | **Allow All Network Access** / **Defaults + Team Allowlist** / **Team Allowlist Only**; no policy = allow-all | Ent | | Team Setup (Cursor: "Team Setup Scripts") | install-script manifests on every computer — see [[concepts/private-networks]] | Ent | | Team Secrets | up to 100 per team, 32 KB per value, 96 KB total; reserved names such as `PATH`, `HOME`, and anything starting with `SAND_` or `LD_` are rejected; dashboard only, not via Admin API | Ent | | Manage Bot Computers (Cursor: "Manage Member Computers and Data") | Recreate / Terminate (Cursor adds Delete and delete-all-data); **Terminate Inactive Computers** after 30 days unused, off by default | Ent | | Action Recording | off by default; exported via OpenTelemetry tagged `cursor.surface=grok_bot`; not shown on the Audit Log page | Ent | | Audit logs | control-plane events (Bot creation, access changes, Team Setup manifests, MCP auth, Slack links, routines) | Ent | | OpenTelemetry Export | **Team Settings → OpenTelemetry Export** | Ent | **Team Secrets behaviour:** read by Setup and Check Scripts as env vars (e.g. `"$MY_LICENSE_KEY"`); exist only in the script's process; script output is redacted; a computer whose Team Setup comes from two or more teams gets *no* secrets; changes apply next time scripts run (start, periodic refresh, or recreate). **Group settings** (Ent): open a group under Members > Groups → **Grok Bot** tab. Groups only *widen* the team baseline; a member in several groups gets the most permissive result. **Network is the exception**: a group policy replaces the team's (looser or tighter), **Lock for All Groups** keeps the team policy for everyone, and multi-group members get the most restrictive. Sections: Bot Capabilities, Network, Group Rules, Setup Scripts (no group-level secrets). **Admin API:** `https://cursor.com/docs/account/teams/admin-api#grok-bot` covers enable, capabilities, Enforce Auto-review, group access, network policy, team rules and setup scripts. Cursor's page adds that bulk recreate/terminate/delete uses the Organization API with an Organization API key carrying the `admin:*` scope. No spend cap specific to Grok Bot exists today. ## When To Use Read this page before enabling Grok Bot for an organization or when a member asks why a feature is missing (many controls are Enterprise-only and simply don't appear on self-serve Teams). For the ordered rollout, use [[syntheses/enterprise-rollout-guide]]. Both pages' recommended admin baseline: configure Network Controls; audit connector policy before enabling; turn on Enforce Auto-review before relying on team rules; set Execution on Local Computer to **Never allow** unless needed; disable Cloud Agent spawning if unused; keep public template sharing off; add team Auto-review rules (production deploys, external email, payments, accepting legal terms as **Ask first**); gate sign-in to managed devices via your IdP. ## Risks & Pitfalls - **Privacy Mode (Legacy) blocks Grok Bot entirely** and, per the docs.x.ai page only, also blocks conversation content export. - **Disable ≠ deprovision.** Turning Grok Bot off or terminating a computer does not remove access; remove the member from the team or group and revoke IdP sessions. - **Team admin rights are not enough** for bulk computer management (xAI: organization admins only, because one computer spans every team a member belongs to). - **Auto-review rules need enforcement**: turning enforcement off stops applying team rules. Cursor notes Team Bots in chats where nobody can answer run *without* Auto-review unless it is enforced. - **An allowlist mode does not restrict MCP servers** by itself (Cursor page): turn on **Apply to MCP servers** under Network (off by default) and add connector hosts such as `api.github.com`, `api.x.com`, `api.origin.cursor.com`. - Any permitted connector is available to every Bot a member runs. **Divergences between hosts (as of fetch 2026-10-21):** Cursor's page has Inference Hooks, Manage Team Bots, Apply to MCP servers, and Delete/delete-all-data for computers, which xAI's lacks; xAI's page has a **Conversation content export** section (prompts/responses, opt-in **Tool I/O** for MCP calls, redaction to `[REDACTED: ...]`) that Cursor's teams page omits; xAI says Conversation Insights is available "on Enterprise teams where it has rolled out", Cursor says "on all Enterprise teams". ## Related Concepts - [[concepts/identity-and-access]], [[concepts/enterprise-deployment]], [[concepts/private-networks]], [[concepts/proxies]] - [[concepts/approvals]], [[concepts/privacy-and-data]], [[syntheses/security-model]] ## Sources - raw/web_community-teams-and-enterprises-md.md — https://docs.x.ai/grok-bot/teams-and-enterprises.md - raw/web_community-teams-md.md — https://cursor.com/docs/grok-bot/teams.md --- title: "Use Cases and Starter Prompts" type: concept tags: [grok-bot, use-cases, prompts, roles, starter-prompts] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-use-cases-md-2.md", "raw/web_community-use-cases-md.md", "raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md"] --- # Use Cases and Starter Prompts ## Definition The official use-case pages (docs.x.ai `use-cases.md` and cursor.com `use-cases.md`) describe eight Bot **roles**. Each role owns a repeatable outcome rather than a loose category of questions. For each role the pages list what it owns, what to connect, a starter prompt, and a follow-on tip. cursor.com points out that **every starter prompt deliberately ends at a review point**: sending, publishing, purchasing, and production changes stay behind approval. The prompts below are quoted from docs.x.ai. cursor.com's prompts match apart from contractions ("don't" for "do not", etc.) and minor omissions such as "(ICP)", "(CAC)", "potential", "flamegraphs" and "relevant console or network notes", noted where they appear; cursor.com's "Connect" lists are also shorter, and the Connect lists below are docs.x.ai's. ## How It Works Recommended pattern: start each role with read-and-prepare work, review the result, and only then add approved actions or a routine. ### Sales Outbound **Owns:** account research, contact prioritization, review-ready outreach. **Connect:** CRM, product-intent sources, company websites, email, and professional networks as permitted by their terms. > Research the 25 accounts in this CRM view. Score them against our ideal customer profile (ICP) and recent intent, identify up to three relevant contacts per account, and draft email and LinkedIn outreach in the style examples attached. Skip anyone already in an active sequence. Return a review list; do not send or enroll anyone. Next: once the output is reliable, add a nightly research routine that stops at the review list. ### Talent Scout **Owns:** sourcing, candidate research, outreach drafts, scheduling preparation. **Connect:** applicant tracking system (ATS), approved sourcing tools, email, calendar. > For this role description, find 20 potential candidates who meet the must-have criteria. Exclude anyone already in our ATS, explain the evidence for each match, and draft personalized outreach in my voice. Do not contact anyone. Next: add approvals before external outreach, and have the Bot respect candidate privacy, regional requirements, and source terms. ### Paid Media **Owns:** campaign monitoring and budget recommendations. **Connect:** advertising platforms, analytics, budget spreadsheet, Slack. > Pull current spend and performance by campaign. Compare it with the monthly budget and target customer acquisition cost (CAC), then recommend reallocations with the supporting numbers. Draft a Slack update for the growth team. Do not change budgets or send the message. Next: keep campaign changes behind approval even after the analysis becomes a routine. ### Expense Manager **Owns:** weekly expense reconciliation and missing-information follow-up. **Connect:** expense system, email, shared drive, finance spreadsheets. > Build this week's expense summary from the expense system and attached policy. Match receipts from the finance inbox, flag missing categories or policy exceptions, and draft one follow-up per owner. Return the summary and drafts; do not send messages or change reimbursements. Next: require a policy citation on every exception and totals that reconcile back to the source. ### Product Performance **Owns:** targeted performance investigations with evidence. **Connect:** observability, analytics, incident tooling, source-control links. > Investigate the checkout latency increase since yesterday's release. Review dashboards, traces, and flamegraphs; identify the highest-confidence hotspot; and return a short write-up with screenshots and direct links. Separate facts from hypotheses. Do not change alerts or production settings. (cursor.com's version omits "flamegraphs".) Next: use a routine for a recurring health report, not for unsupervised production changes. ### Bug Reproduction **Owns:** turning reports into reliable reproduction packs. **Connect:** issue tracker, staging environment, browser, network tools. > Read this bug report and reproduce it in staging using a fresh test account. Return exact steps, expected and actual behavior, screenshots, browser and OS details, relevant console or network notes, and a minimal test case if possible. Do not use production customer data. (cursor.com's version omits "relevant console or network notes".) Next: provide approved test credentials through a secure handoff, not in chat. ### Account Health **Owns:** risk and expansion signals across a customer portfolio. **Connect:** CRM, product usage, support, billing, customer-success notes. > Review the accounts in this portfolio. Combine recent usage, support escalations, renewal timing, and stakeholder activity into a ranked watch list. For each account, include the evidence, why it matters, and a suggested next step. Do not contact customers or edit the CRM. Next: define risk thresholds in the Bot description so the weekly result stays consistent. ### Chief of Staff **Owns:** a source-linked digest of what changed and what needs attention. **Connect:** Slack, email, calendar, meeting notes, planning documents. > Review activity since yesterday across my approved channels, inbox, calendar, and meeting notes. Return only items that map to the priorities in this document. For each item, include the source, why it matters, the proposed next step, and whether I owe a decision. Do not send messages or change meetings. Next: mark what was useful and what was noise, then schedule the digest for a time when you can review it. ## Key Parameters **Turning a role into a durable Bot** (docs.x.ai lists seven steps; cursor.com folds them into six): 1. Put the job, source systems, output format, and standing boundaries in the Bot description. 2. Run one real task with a safe scope. 3. Correct the result until it is reviewable (cursor.com: "without you"). 4. Save the successful process as a skill. 5. Test it on a second input. 6. Create a routine only once retries and failure cases are defined. 7. Keep consequential external actions behind approval. Note: cursor.com elsewhere says personal Bots have no description field and that lasting rules should be given in chat (see [[concepts/bots]]). cursor.com's own use-cases step 1 nonetheless says "Bot's description", so this is an internal cursor.com inconsistency. **Field reports (third-party, medium confidence):** GIGAZINE's launch coverage (Aug 12, 2026) reports that at SpaceXAI a sales Bot enters call records into the CRM and drafts follow-up emails, while an engineering Bot reproduces bugs on product screens and files tickets. It also quotes staff using Bots for calendar management, shipping code, and a fleet of specialized "thought partner" Bots for product-growth work. ## When To Use Pick a role whose output you can check, such as a review list, watch list, digest, or repro pack. Avoid unbounded "general helper" roles. Move from a one-off task to a skill and then a routine only after the output is reliable (see [[concepts/skills-routines-and-automations]]). ## Risks & Pitfalls - Don't drop the "do not send / do not change" clauses when you adapt a prompt. They are the review points that keep external actions behind [[concepts/approvals]]. - Credentials for test or staging accounts belong in a secure handoff, never in chat. - Automating too early: create routines only after retries and failure cases are defined. - Sourcing and outreach roles carry privacy, regional, and terms-of-service obligations that the docs explicitly call out. ## Related Concepts - [[concepts/bots]] - [[concepts/skills-routines-and-automations]] - [[concepts/files-and-results]] - [[concepts/apps-and-connectors]] - [[syntheses/getting-started]] ## Sources - `raw/web_community-use-cases-md-2.md`: https://docs.x.ai/grok-bot/use-cases.md - `raw/web_community-use-cases-md.md`: https://cursor.com/docs/grok-bot/use-cases.md - `raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md`: https://gigazine.net/gsc_news/en/20260812-spacexai-grok-bot/ (third-party) --- title: "Grok Bot" type: entity tags: [grok-bot, product, xai, cursor, spacexsi, beta] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-overview-md.md", "raw/web_community-grok-bot-md.md", "raw/web_community-faq-md.md", "raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md", "raw/web_community-release-notes-md.md", "raw/web_community-deployment-md.md", "raw/web_community-security-md.md", "raw/web_community-security-md-2.md", "raw/web_community-bots-md.md", "raw/web_community-get-started-md.md", "raw/web_community-get-started-md-2.md", "raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md", "raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md", "raw/web_community-official-grok-bot-breakdown-2026.md"] --- # Grok Bot ## Overview Grok Bot is a product for creating **Bots**: persistent, named AI teammates. Each Bot works on a cloud computer with a browser, filesystem and terminal, and finishes tasks in your real tools. You direct Bots by messaging them, and they ask you when something needs approval. **Who makes and ships it.** The official material splits the product across two companies: - **xAI side.** The xAI website hosts the launch and plan news (x.ai/news). The page title reads "SpaceXSI", and the xAI developer release notes also use "SpaceXSI". The docs at docs.x.ai/grok-bot also say that shared Bots are created by other users, "not by SpaceXSI". - **Cursor side.** Cursor provides the account, sign-in, privacy settings and the cloud: "The computers Bots work on run in Cursor's cloud." The same docs are also published at cursor.com/docs/grok-bot. - **Third-party naming.** Gigazine says "SpaceXAI". Vellum and Davies Meyer say "xAI". According to Gigazine, SpaceXAI's David Pan wrote that "Most of Cursor now uses it every day" at launch. **Status and dates** - **Launch.** It launched in beta on **August 11** (x.ai news, 2026-08-26). Gigazine and Davies Meyer call it an "early beta" released 2026-08-11. - **Release notes.** The xAI developer release notes list "Grok Bot is now available" under **August**. - **Plan expansion.** Access was expanded on **Aug 26, 2026**. - **Today.** The x.ai news post calls it a beta. The current docs (fetched 2026-10-21) don't label the product as a whole as beta. Only individual features, such as inference hooks, are marked "beta". Timeline: [[summaries/release-history]]. ## Characteristics | Characteristic | Detail (official unless attributed) | |---|---| | Unit of work | Named Bots, each with one job. Context compounds per Bot: memory, files, browser sessions and preferences | | Computer | One persistent hosted computer per user, shared by all of that user's Bots. The Cursor deployment page says it "runs Linux in Cursor's cloud and isn't enrolled in MDM" and is updated by Cursor | | Concurrency | Bots work in parallel. Each Bot gets its own screen, and one Bot runs one computer-use task at a time | | Integrations | Connectors where available, computer use for everything else | | Automation | Skills (how to do a task) and routines (a scheduled or event-triggered workflow for one Bot). A Bot can learn a workflow from a demonstration | | Models | No customer-facing model picker. The Cursor security page says "Grok Bot manages model selection" and that each request goes to a first-party xAI model or a third-party model from a sub-processor. The x.ai security page says "Cursor manages model selection" | | Desktop clients | macOS (Apple silicon and Intel), Windows (x64 and Arm64), Linux (x64 and Arm64) | | Mobile clients | The x.ai docs list iPhone/iPad (iOS/iPadOS 18 or later) and Android 9 or later. The Cursor overview lists iPhone only, though Cursor's deployment page lists Android 9 or later (Google Play) as well. See [[concepts/mobile-app]] | | Plans | Included with paid individual Cursor plans, Cursor Teams and linked SuperGrok tiers. Cursor docs add Grok Business seats. Usage resets weekly. See [[concepts/plans-and-access]] | | Account | Uses your Cursor account. There is no separate Grok Bot login (Cursor get-started) | **Third-party claims (medium/low confidence; not in the official docs):** - **Vellum's breakdown:** - Calls the computer an "account-scoped persistent cloud VM". - Says custom MCP servers must be publicly reachable, or tunneled. - Lists built-in connectors: Gmail, Google Calendar, Google Drive, OneDrive, Outlook Mail and Calendar, Microsoft Teams, SharePoint and Salesforce. - States a **14-day free trial, then $200 per month** standalone price, with no self-hosting and no way to use non-xAI models. - The official pages in this corpus don't confirm the standalone price or the trial. - No source in the corpus gives VM hardware specs (CPU/RAM/disk). - **Gigazine** (2026-08-12) reported the launch-time plan prices: SuperGrok Heavy $300/month, Cursor Ultra $200/month, and Cursor Teams Premium $120 per user per month. - **Davies Meyer** has a section headed "The model underneath: Grok 4.6", but its text only says Grok Bot "lands in the same month as Grok 4.6". The official Grok Bot docs don't name an underlying model. ## How to Use 1. **Check eligibility** on [[concepts/plans-and-access]]. 2. **Install the desktop app.** Get it from [x.ai/bot](https://x.ai/bot), or from the **Download Grok Bot** row on cursor.com/dashboard/bot. 3. **Sign in.** Use your Cursor account. Link SuperGrok if that's where your access comes from. 4. **Create a Bot** with a short name, one primary job and a description of how it should work. 5. **Message it a task.** State the outcome, sources, constraints, deliverable and review point. 6. **Approve, take over, refine.** Approve or take over when it asks, then turn stable processes into skills or routines. The full sequence is in [[syntheses/getting-started]]. You can also start work by tagging @bot on X ([[concepts/tag-bot-on-x]]). ## Related Entities - [[syntheses/overview]]: what the product is and where to go next - [[concepts/bots]], [[concepts/team-bots]], [[concepts/cloud-computer]], [[concepts/apps-and-connectors]] - [[concepts/teams-and-enterprises]], [[concepts/enterprise-deployment]] - [[summaries/release-history]], [[summaries/docs-map]] ## Sources - raw/web_community-overview-md.md — https://docs.x.ai/grok-bot/overview.md - raw/web_community-grok-bot-md.md — https://cursor.com/docs/grok-bot.md - raw/web_community-faq-md.md — https://docs.x.ai/grok-bot/faq.md - raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md — https://x.ai/news/grok-bot-more-plans - raw/web_community-release-notes-md.md — https://docs.x.ai/developers/release-notes.md (Grok Bot entry only) - raw/web_community-deployment-md.md — https://cursor.com/docs/grok-bot/deployment.md - raw/web_community-security-md.md — https://cursor.com/docs/grok-bot/security.md - raw/web_community-security-md-2.md — https://docs.x.ai/grok-bot/security.md - raw/web_community-bots-md.md — https://docs.x.ai/grok-bot/bots.md - raw/web_community-get-started-md.md — https://cursor.com/docs/grok-bot/get-started.md - raw/web_community-get-started-md-2.md — https://docs.x.ai/grok-bot/get-started.md - raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md — https://gigazine.net/gsc_news/en/20260812-spacexai-grok-bot/ (third-party) - raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md — https://ai-solutions.daviesmeyer.com/en/blog/grok-bot-ki-teammates-xai (third-party) - raw/web_community-official-grok-bot-breakdown-2026.md — https://www.vellum.ai/blog/official-grok-bot-breakdown (third-party) --- title: "Activity Log" type: log --- # Activity Log Append-only record of all wiki changes. ## Format Each entry follows this format: ``` ### YYYY-MM-DD HH:MM — [Action Type] - **Source/Trigger**: what initiated the action - **Pages created**: list of new pages - **Pages updated**: list of updated pages - **Notes**: any contradictions flagged, decisions made ``` --- ### 2026-04-08 00:00 — Setup - **Source/Trigger**: Repository initialized - **Pages created**: index.md, log.md, dashboard.md, analytics.md, flashcards.md - **Pages updated**: none - **Notes**: Empty knowledge base ready for first source ingestion ### 2026-10-21 — Initial build - **Source/Trigger**: new topic wiki on Grok Bot, gathered from 37 official doc pages across both hosts + news, release notes and 3 coverage pieces - **Pages created**: 28 (5 syntheses · 20 concepts · 1 entities · 2 summaries); see [[index]] - **Pages updated**: none - **Notes**: No product version is published, so freshness = fetch date. Main cross-host conflicts flagged on pages: inference hooks (xAI security page says DLP hooks are not available; Cursor documents an Enterprise beta), model allowlist, auto-review rule storage, usage reporting, platform list. Official xAI pages spell the company 'SpaceXSI'; pages keep each source's spelling. --- title: "Grok Bot Docs Map" type: summary tags: [grok-bot, docs-map, catalog, sources, divergences] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-approvals-security-and-privacy-md.md", "raw/web_community-bots-md.md", "raw/web_community-chat-and-collaboration-md.md", "raw/web_community-computer-and-apps-md.md", "raw/web_community-computers-md-2.md", "raw/web_community-computers-md.md", "raw/web_community-deployment-md-2.md", "raw/web_community-deployment-md.md", "raw/web_community-faq-md.md", "raw/web_community-files-and-results-md.md", "raw/web_community-get-started-md-2.md", "raw/web_community-get-started-md.md", "raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md", "raw/web_community-grok-bot-md.md", "raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md", "raw/web_community-identity-and-access-md.md", "raw/web_community-identity-md.md", "raw/web_community-inference-hooks-md.md", "raw/web_community-mobile-md.md", "raw/web_community-official-grok-bot-breakdown-2026.md", "raw/web_community-overview-md.md", "raw/web_community-private-networks-md-2.md", "raw/web_community-private-networks-md.md", "raw/web_community-proxies-md-2.md", "raw/web_community-proxies-md.md", "raw/web_community-release-notes-md.md", "raw/web_community-security-faq-md-2.md", "raw/web_community-security-faq-md.md", "raw/web_community-security-md-2.md", "raw/web_community-security-md.md", "raw/web_community-settings-and-notifications-md.md", "raw/web_community-settings-md.md", "raw/web_community-skills-routines-and-automations-md.md", "raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md", "raw/web_community-tag-on-x-md.md", "raw/web_community-team-bots-md.md", "raw/web_community-teams-and-enterprises-md.md", "raw/web_community-teams-md.md", "raw/web_community-troubleshooting-md.md", "raw/web_community-use-cases-md-2.md", "raw/web_community-use-cases-md.md", "raw/web_community-work-md.md"] --- # Grok Bot Docs Map ## Key Points This wiki has 42 raw sources: - **Official docs on two hosts.** The same product is documented at docs.x.ai/grok-bot/* (23 pages) and cursor.com/docs/grok-bot* (14 pages). File names don't reveal the host. A `-2` suffix just means a second fetch of the same slug, so this map uses each file's `source_url`. - **One x.ai news post and one release-notes page.** Only the Grok Bot entry in the release notes is in scope. - **Three third-party articles.** All were fetched 2026-10-21. ### Catalog (one row per raw file) | Raw file | Host | Page | Topic | Covered by | |---|---|---|---|---| | raw/web_community-overview-md.md | docs.x.ai | /grok-bot/overview.md | What Grok Bot is, the shared computer, tagging @bot, FAQ | [[syntheses/overview]], [[entities/grok-bot]] | | raw/web_community-grok-bot-md.md | cursor.com | /docs/grok-bot.md | Cursor's overview (pairs with x.ai overview) | [[syntheses/overview]], [[entities/grok-bot]] | | raw/web_community-get-started-md-2.md | docs.x.ai | /grok-bot/get-started.md | Install (per-OS steps), sign in, first Bot, first task | [[syntheses/getting-started]] | | raw/web_community-get-started-md.md | cursor.com | /docs/grok-bot/get-started.md | Install, sign in, first Bot, iOS section | [[syntheses/getting-started]] | | raw/web_community-mobile-md.md | docs.x.ai | /grok-bot/mobile.md | iPhone/iPad/Android app | [[concepts/mobile-app]] | | raw/web_community-faq-md.md | docs.x.ai | /grok-bot/faq.md | General FAQ: cost, platforms, memory, sharing | [[syntheses/overview]], [[concepts/plans-and-access]] | | raw/web_community-use-cases-md-2.md | docs.x.ai | /grok-bot/use-cases.md | 8 role templates with starter prompts | [[concepts/use-cases]] | | raw/web_community-use-cases-md.md | cursor.com | /docs/grok-bot/use-cases.md | Same 8 roles | [[concepts/use-cases]] | | raw/web_community-bots-md.md | docs.x.ai | /grok-bot/bots.md | Create, edit, pin, duplicate, share, delete Bots; Marketplace | [[concepts/bots]] | | raw/web_community-team-bots-md.md | docs.x.ai | /grok-bot/team-bots.md | Shared team Bots, Slack, usage attribution | [[concepts/team-bots]] | | raw/web_community-chat-and-collaboration-md.md | docs.x.ai | /grok-bot/chat-and-collaboration.md | Messaging, group chats, handoffs, shortcuts | [[concepts/messaging-and-collaboration]] | | raw/web_community-tag-on-x-md.md | docs.x.ai | /grok-bot/tag-on-x.md | Tag @bot on X, regions | [[concepts/tag-bot-on-x]] | | raw/web_community-files-and-results-md.md | docs.x.ai | /grok-bot/files-and-results.md | Attachments, previews, evidence | [[concepts/files-and-results]] | | raw/web_community-computer-and-apps-md.md | docs.x.ai | /grok-bot/computer-and-apps.md | Shared computer, takeover, connecting apps, X search | [[concepts/cloud-computer]], [[concepts/apps-and-connectors]] | | raw/web_community-skills-routines-and-automations-md.md | docs.x.ai | /grok-bot/skills-routines-and-automations.md | Skills, Teach a task, routines, event triggers | [[concepts/skills-routines-and-automations]] | | raw/web_community-work-md.md | cursor.com | /docs/grok-bot/work.md | One page combining Bots, messaging, computer and apps, skills and routines | [[concepts/bots]], [[concepts/messaging-and-collaboration]], [[concepts/cloud-computer]], [[concepts/skills-routines-and-automations]] | | raw/web_community-settings-and-notifications-md.md | docs.x.ai | /grok-bot/settings-and-notifications.md | Settings panes, attention states, notification routing | [[concepts/settings-and-notifications]] | | raw/web_community-settings-md.md | cursor.com | /docs/grok-bot/settings.md | Settings (General, Computer, Usage & Billing, Updates) | [[concepts/settings-and-notifications]] | | raw/web_community-approvals-security-and-privacy-md.md | docs.x.ai | /grok-bot/approvals-security-and-privacy.md | User-level approvals, Auto Review, takeover, data settings | [[concepts/approvals]], [[concepts/privacy-and-data]] | | raw/web_community-security-md-2.md | docs.x.ai | /grok-bot/security.md | Org security: network, egress, logging, retention, models | [[syntheses/security-model]], [[concepts/privacy-and-data]] | | raw/web_community-security-md.md | cursor.com | /docs/grok-bot/security.md | Same, plus architecture, MCP network rules, inference hooks | [[syntheses/security-model]], [[concepts/privacy-and-data]] | | raw/web_community-security-faq-md-2.md | docs.x.ai | /grok-bot/security-faq.md | Security review Q&A | [[syntheses/security-model]] | | raw/web_community-security-faq-md.md | cursor.com | /docs/grok-bot/security-faq.md | Security review Q&A | [[syntheses/security-model]] | | raw/web_community-inference-hooks-md.md | cursor.com | /docs/grok-bot/inference-hooks.md | Enterprise-beta model-request hooks | [[concepts/inference-hooks]] | | raw/web_community-teams-and-enterprises-md.md | docs.x.ai | /grok-bot/teams-and-enterprises.md | Plans table, admin controls, rollout | [[concepts/teams-and-enterprises]], [[syntheses/enterprise-rollout-guide]] | | raw/web_community-teams-md.md | cursor.com | /docs/grok-bot/teams.md | Same topic | [[concepts/teams-and-enterprises]], [[syntheses/enterprise-rollout-guide]] | | raw/web_community-deployment-md-2.md | docs.x.ai | /grok-bot/deployment.md | Installers, release feed, MDM push | [[concepts/enterprise-deployment]] | | raw/web_community-deployment-md.md | cursor.com | /docs/grok-bot/deployment.md | Same; rollout steps given as H3 sections | [[concepts/enterprise-deployment]] | | raw/web_community-identity-and-access-md.md | docs.x.ai | /grok-bot/identity-and-access.md | Okta / Entra ID assignment and device trust | [[concepts/identity-and-access]] | | raw/web_community-identity-md.md | cursor.com | /docs/grok-bot/identity.md | Same | [[concepts/identity-and-access]] | | raw/web_community-computers-md-2.md | docs.x.ai | /grok-bot/computers.md | Admin: recreate/terminate computers, auto-terminate, API | [[concepts/cloud-computer]], [[concepts/teams-and-enterprises]] | | raw/web_community-computers-md.md | cursor.com | /docs/grok-bot/computers.md | Same | [[concepts/cloud-computer]], [[concepts/teams-and-enterprises]] | | raw/web_community-private-networks-md-2.md | docs.x.ai | /grok-bot/private-networks.md | Team Setup, networking clients | [[concepts/private-networks]] | | raw/web_community-private-networks-md.md | cursor.com | /docs/grok-bot/private-networks.md | Team Setup Scripts | [[concepts/private-networks]] | | raw/web_community-proxies-md-2.md | docs.x.ai | /grok-bot/proxies.md | Gateways/proxies | [[concepts/proxies]] | | raw/web_community-proxies-md.md | cursor.com | /docs/grok-bot/proxies.md | Same | [[concepts/proxies]] | | raw/web_community-troubleshooting-md.md | docs.x.ai | /grok-bot/troubleshooting.md | Symptom-by-symptom fixes | [[syntheses/troubleshooting]] | | raw/web_community-release-notes-md.md | docs.x.ai | /developers/release-notes.md | Grok Bot "now available" entry (August) | [[summaries/release-history]] | | raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md | x.ai | /news/grok-bot-more-plans | 2026-08-26 plan expansion | [[concepts/plans-and-access]], [[summaries/release-history]] | | raw/web_community-official-grok-bot-breakdown-2026.md | vellum.ai | /blog/official-grok-bot-breakdown | Third-party review (competitor) | [[entities/grok-bot]] | | raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md | gigazine.net | /gsc_news/en/20260812-spacexai-grok-bot/ | Third-party launch report | [[summaries/release-history]], [[entities/grok-bot]] | | raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md | daviesmeyer.com | /en/blog/grok-bot-ki-teammates-xai | Third-party marketing-team take | [[entities/grok-bot]] | ### Topics on only one host - **Only on docs.x.ai:** - mobile - FAQ - bots - team-bots (Cursor links to a help-center page that isn't in the corpus) - chat-and-collaboration - tag-on-x - files-and-results - computer-and-apps - skills-routines-and-automations - approvals-security-and-privacy - troubleshooting Cursor covers several of these in one combined page, work.md. - **Only on cursor.com:** - inference-hooks (the x.ai docs don't mention inference hooks at all) - work.md as a combined page ### Substantive divergences between paired pages (x.ai vs Cursor, both fetched 2026-10-21) | Pair | Divergence | |---|---| | overview vs grok-bot.md | Platforms: x.ai lists "macOS, Windows, Linux, iOS, and Android" and a mobile app on "iPhone, iPad, and Android". Cursor lists "macOS, Windows, Linux, and iOS" and a mobile app on "iPhone". Plans: only Cursor adds "a seat on a self-serve Grok Business plan". x.ai mentions voice input and tagging @bot on X. Cursor mentions Team Bots and a Deployment next step | | get-started | Download: x.ai sends you to x.ai/bot with per-OS steps. Cursor sends you to the **Download Grok Bot** row on cursor.com/dashboard/bot. First Bot: x.ai shows **Meet a future teammate** / **Create your own**. Cursor auto-creates a Bot named **Grok Bot**. New Bot: **New → Create new Bot** (x.ai) vs **New Bot** (+) (Cursor). Takeover: **Take over** / **I'm done** vs **Open computer**. Legacy privacy: "move to a supported Cursor data setting" vs "prompted to switch to Privacy Mode". Only Cursor mentions Grok Business and has a dedicated iOS section; Cursor's page never mentions Android, while x.ai's mentions iPhone and Android (dictation, Link Grok Account) and links its mobile page | | security | Cursor adds **Architecture**, **MCP servers and the network policy** (an allowlist mode alone doesn't restrict MCP servers; built-in connectors are checked by host) and **Inference hooks**. Models: x.ai says "Cursor manages model selection" and that usage analytics show the serving model. Cursor says "Grok Bot manages model selection" and that usage is recorded "as Grok Bot usage rather than by model". Model allowlist: Cursor says it "does not govern Grok Bot"; x.ai says it is "honored by default" but "enforcement is not guaranteed". DLP: x.ai says "Dedicated data loss prevention hooks are not available" while Cursor documents inference hooks. Personal Auto-review rules: x.ai "stored on the current desktop", Cursor "saved to the member's account" (see [[syntheses/security-model]]) | | security-faq | x.ai only: "Does an allowlist turn off web search?" (yes, under **Defaults + Team Allowlist** or **Team Allowlist Only**). Cursor only: "Can our own security tooling check what Grok Bot sends to models?" (inference hooks). Wording: "block a plugin" vs "block an app" | | teams-and-enterprises vs teams | Plans table: only Cursor lists Grok Business seats for individuals. Admin controls: **Bot Capabilities** (x.ai) vs **Agent capabilities** (Cursor). Cursor adds **Manage Team Bots** and **Inference Hooks**. x.ai lists **Conversation content export** (an opt-in OpenTelemetry addition), which Cursor doesn't mention. **Manage Bot Computers** vs **Manage Member Computers and Data** | | computers | Cursor documents **Delete all Grok Bot data** in its own section (at most 100 members, can't be undone). x.ai covers delete actions inline (also up to 100 members), and only x.ai documents **Restart after a settings change** (**Restart all VMs** → **Recreate VMs**). Only x.ai says the API returns per-member results "for operations of up to 1,000 members" | | deployment | Hosted computer: "Cursor's cloud … Updated by Cursor" (Cursor) vs "The cloud … Automatically, in the background" (x.ai). Both have the same rollout order (move off Privacy Mode (Legacy) → network → IdP assignment → enable → configuration → pilot → push to fleet → plan updates). x.ai gives it as a numbered "Rollout checklist", and Cursor gives each step its own H3 section | | private-networks | Name: **Team Setup** (x.ai) vs **Team Setup Scripts** (Cursor). Cursor splits setup into H3 sections (open, manifest, setup script, check script), while x.ai uses a numbered list under "Create the manifest". Cursor places it under **Bot Computers** on the Grok Bot dashboard page, x.ai under the **Grok Bot** page. Both say it's Enterprise only | | settings | Same top-level panes (General, Computer, Usage and billing, Updates). x.ai breaks General into Account / Appearance / Bot / Auto-review and notes "Plugins are not a settings section". Cursor adds **Run commands on your desktop**, **Connected apps** and **Team Setup Scripts** sections | | use-cases | Prompts match except: x.ai's Product Performance prompt says "dashboards, traces, and flamegraphs" (Cursor: no "flamegraphs"), and x.ai's Bug Reproduction prompt lists "relevant console or network notes" (Cursor: omitted). x.ai's Connect lists are longer (e.g. company websites, professional networks, shared drive, source-control links, browser and network tools). x.ai has seven "durable Bot" steps, Cursor six. Only Cursor adds that every prompt ends at a review point | | identity, proxies | Wording only: plugin → connected app, contractions, a few small word swaps (e.g. "egress" vs "traffic"). No differing values found | **General terminology drift:** the x.ai docs say "plugin" or "connector", and the Cursor docs say "connected app" or "app". ## Relevant Concepts [[syntheses/overview]], [[entities/grok-bot]], [[summaries/release-history]], [[syntheses/security-model]], [[concepts/teams-and-enterprises]], [[concepts/apps-and-connectors]] ## Source Metadata - **Fetch date:** all 42 files were fetched 2026-10-21. The official pages carry no publication dates. - **Host rule:** read `source_url` in each file's frontmatter. The x.ai docs start with a section label (for example `#### Manage and protect`). The Cursor docs end with a "Sitemap" footer. - **Third-party files:** - Vellum is a competing product's blog, so treat it as low-to-medium confidence. - Gigazine is dated 2026-08-12. - Davies Meyer is undated. --- title: "Grok Bot Release History" type: summary tags: [grok-bot, release-history, timeline, beta, launch] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-release-notes-md.md", "raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md", "raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md", "raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md", "raw/web_community-official-grok-bot-breakdown-2026.md", "raw/web_community-faq-md.md", "raw/web_community-overview-md.md", "raw/web_community-grok-bot-md.md", "raw/web_community-inference-hooks-md.md", "raw/web_community-tag-on-x-md.md"] --- # Grok Bot Release History ## Key Points ### Timeline | Date | Event | Source (confidence) | |---|---|---| | ~6 weeks before launch (≈ late June 2026) | Grok Bot was "a prototype with a handful of users". By launch, "Most of Cursor now uses it every day" | David Pan's post, quoted by Gigazine (third-party; the date is approximate and derived) | | **2026-08-11** | Launched in beta ("early beta"). Gigazine says the launch plans were SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium, with a business waitlist. At the time, Gigazine said the official docs listed Windows and macOS desktop apps and an iOS app, with "no official desktop app for Linux" (Gigazine's own headline nonetheless lists Linux as compatible) (Linux users could use the Grok Build TUI) and no Android version | x.ai news ("We launched Grok Bot in beta on August 11"); Gigazine 2026-08-12; Davies Meyer | | 2026-08-11 | Elon Musk posted that the beta would widen after fixing "basic issues" and releasing Grok 4.6 "later this week" | Quoted by Gigazine (third-party) | | August 2026 | xAI developer release notes, under **August**: "Grok Bot is now available. Durable AI teammates that work on a persistent cloud computer, with messaging, approvals, connectors, and routines." The same month's notes list Grok 4.6 on the xAI API | docs.x.ai release notes (official; no day given) | | **2026-08-26** | "Grok Bot is now included with more plans": SuperGrok, SuperGrok Plus, SuperGrok Heavy, Cursor Pro, Cursor Pro+, Cursor Ultra, Cursor Teams Plans (Standard and Premium). Usage is separate from Grok and Cursor plan usage. An enterprise waitlist was opened | x.ai news (official) | | Undated (2026) | Vellum's breakdown calls it "early-beta". It describes $200/month standalone with a 14-day trial, desktop on macOS and Windows, and an iPhone companion | Vellum (third-party, vendor competitor) | | As of fetch **2026-10-21** | The docs on both hosts describe a desktop app on macOS, Windows and Linux (x64 and Arm64). The x.ai docs add iPhone/iPad (iOS 18 or later) and Android 9 or later; Cursor's deployment page also lists Android 9 or later, so only Cursor's intro pages omit it. Cursor docs add self-serve Grok Business seats. The x.ai FAQ says "Enterprise access is rolling out". Inference hooks are "in beta for Enterprise teams". Tag @bot on X is available in most countries but not yet in Australia, the UK, the EU, Iceland, Liechtenstein or Norway | x.ai and Cursor docs (official; the docs don't give when each feature shipped) | ### What changed between launch and now - **Plans:** - 2026-08-11 (Gigazine): three top tiers. - 2026-08-26 (x.ai news): all SuperGrok and Cursor Pro/Pro+/Ultra/Teams plans. - By 2026-10-21: "every paid individual Cursor plan", the Teams plan and SuperGrok links; Cursor docs also add Grok Business. - Details: [[concepts/plans-and-access]]. - **Platforms:** launch-time reports say macOS, Windows and iOS. Linux desktop and Android are documented by 2026-10-21. See [[concepts/mobile-app]]. - **Enterprise:** a waitlist in August. As of fetch, the x.ai FAQ says enterprise access is "rolling out", and both teams pages say Enterprise can "Enable Grok Bot for your team from your dashboard". See [[concepts/teams-and-enterprises]]. ### Gaps The corpus has no official changelog dedicated to Grok Bot. The only release-notes entry is the August "now available" line. Feature ship dates for Team Bots, tagging @bot, inference hooks, Linux builds and Android are not documented. ## Relevant Concepts [[entities/grok-bot]], [[concepts/plans-and-access]], [[concepts/mobile-app]], [[concepts/tag-bot-on-x]], [[concepts/inference-hooks]], [[summaries/docs-map]] ## Source Metadata | File | Host / URL | Date | Type | |---|---|---|---| | raw/web_community-release-notes-md.md | https://docs.x.ai/developers/release-notes.md | August section | Official (only the Grok Bot entry is in scope) | | raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md | https://x.ai/news/grok-bot-more-plans | Aug 26, 2026 | Official news | | raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md | https://gigazine.net/gsc_news/en/20260812-spacexai-grok-bot/ | Aug 12, 2026 | Third-party news | | raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md | https://ai-solutions.daviesmeyer.com/en/blog/grok-bot-ki-teammates-xai | undated | Third-party blog | | raw/web_community-official-grok-bot-breakdown-2026.md | https://www.vellum.ai/blog/official-grok-bot-breakdown | undated (2026) | Third-party, competitor | | raw/web_community-faq-md.md | https://docs.x.ai/grok-bot/faq.md | fetched 2026-10-21 | Official docs | | raw/web_community-overview-md.md | https://docs.x.ai/grok-bot/overview.md | fetched 2026-10-21 | Official docs | | raw/web_community-grok-bot-md.md | https://cursor.com/docs/grok-bot.md | fetched 2026-10-21 | Official docs | | raw/web_community-inference-hooks-md.md | https://cursor.com/docs/grok-bot/inference-hooks.md | fetched 2026-10-21 | Official docs | | raw/web_community-tag-on-x-md.md | https://docs.x.ai/grok-bot/tag-on-x.md | fetched 2026-10-21 | Official docs | --- title: "Enterprise Rollout Guide" type: synthesis tags: [rollout, admin, checklist, enterprise, comparison, xai-vs-cursor] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-teams-and-enterprises-md.md", "raw/web_community-teams-md.md", "raw/web_community-deployment-md-2.md", "raw/web_community-deployment-md.md", "raw/web_community-identity-and-access-md.md", "raw/web_community-identity-md.md", "raw/web_community-private-networks-md-2.md", "raw/web_community-private-networks-md.md", "raw/web_community-proxies-md-2.md", "raw/web_community-proxies-md.md", "raw/web_community-security-md.md", "raw/web_community-security-md-2.md"] --- # Enterprise Rollout Guide An ordered admin checklist assembled from the official deployment checklist and the teams, identity, network and proxy pages on both hosts (`docs.x.ai/grok-bot/*` and `cursor.com/docs/grok-bot/*`). Step order follows the official "Rollout checklist" on the deployment pages; the network-from-the-computer step (5), the policies step (6) and monitoring (10), plus the step groupings, are this wiki's arrangement. ## Comparison ### Ordered checklist | # | Step | Key actions | Page | | --- | --- | --- | --- | | 1 | **Plan** | Confirm plan: Teams = on for everyone, no off switch; Enterprise = admin enables. Move off **Privacy Mode (Legacy)** (blocks Grok Bot entirely). Note Enterprise-only controls. | [[concepts/teams-and-enterprises]], [[concepts/plans-and-access]] | | 2 | **Clear the gateway** | Allow `*.cursor.sh`, `*.cursor-cdn.com`, `*.cursorapi.com`, `*.cursorvm.com`, `*.*.cursorvm.com`, `cursor.com`, `downloads.cursor.com`; exempt from TLS inspection and buffering on every profile incl. off-network. | [[concepts/proxies]] | | 3 | **Identity** | Widen assignment of the existing Cursor SSO app (and SCIM app, Enterprise) in Okta/Entra; add a scoped rule so IdP apps open in the Linux computer browser (Okta **Other Desktop**; Entra Linux-only MFA policy). | [[concepts/identity-and-access]] | | 4 | **Turn on and scope** | Enterprise: **Enable Grok Bot** + **Manage Group Access**. Invite with **Invite Team**. | [[concepts/teams-and-enterprises]] | | 5 | **Network from the computer** | Enterprise: set **Grok Bot Network Access** mode; if allowlisting, decide on **Apply to MCP servers** (Cursor docs). Optionally Team Setup + Team Secrets for Tailscale/Cloudflare Tunnel. | [[concepts/private-networks]], [[syntheses/security-model]] | | 6 | **Policies and approvals** | Apply the recommended configuration: connector policy audit, **Enforce Auto-review**, team Auto-review rules, Execution on Local Computer = **Never allow** unless needed, Cloud Agents off if unused, public template sharing off, Team Rules. | [[concepts/approvals]], [[concepts/teams-and-enterprises]] | | 7 | **Pilot** | Install the chosen version on a few devices per platform, sign in, confirm the app connects to the hosted computer; validate Team Setup scripts on a pilot computer. | [[concepts/enterprise-deployment]] | | 8 | **Deploy** | Push the pinned installer (Windows: `Grok_Bot_{version}_Setup.exe /S`, user context, not SYSTEM). | [[concepts/enterprise-deployment]] | | 9 | **Plan updates** | Pick a cadence; for Linux choose repo upgrades vs. shipping your own packages (`apt-mark hold grok-bot` or repo off). | [[concepts/enterprise-deployment]] | | 10 | **Monitor** | Enterprise: Action Recording → OpenTelemetry Export (`cursor.surface=grok_bot`); audit logs (`grok_bot_vm`, `grok_bot_vm_bulk`); Conversation Insights; optional inference hooks (beta). | [[concepts/teams-and-enterprises]], [[concepts/inference-hooks]] | ### xAI vs Cursor enterprise docs coverage (as of fetch 2026-10-21) | Topic | docs.x.ai | cursor.com | Notes | | --- | --- | --- | --- | | Teams page | `teams-and-enterprises` | `teams` | Same structure | | Individuals availability | Paid Cursor plan or SuperGrok link | Adds self-serve **Grok Business** seat | Cursor broader | | Inference Hooks control | Absent; security page says dedicated DLP hooks "are not available" | Present (Enterprise, beta) | Real divergence | | Apply to MCP servers | Absent | Present, off by default | Cursor only | | Manage Team Bots | Absent | Present | Cursor only | | Computer management | Teams page: Recreate / Terminate; the x.ai computers page also documents Delete, Delete All Grok Bot Data (up to 100 members) and the Organization API with `admin:*` scope | Delete and delete-all-data on the teams page itself; Organization API with `admin:*` scope | Same capabilities; Cursor's teams page is broader | | Conversation content export | Documented on teams page | Not on teams page | xAI only (in this corpus) | | Conversation Insights | "where it has rolled out" | "all Enterprise teams" | Availability wording differs | | Agent Email detail | Brief | `mail.grokbot.com`, drafts until Send | Cursor richer | | Deployment values | Identical feed, download URLs, paths | Identical (App Store link carries a `grok-bot/` path segment: `/app/grok-bot/id6794501026`) | Wording only | | Identity | Says "plugin sign-in" | Says "connected-app sign-in" | Terminology only | | Private networks | Desktop route at **Settings → Computer → Network** | **Settings > Computer**, under **Your computer** | UI path differs | | Proxies | Toggle named "Route egress through this desktop" | "Route traffic through this computer" | Naming only | ## Analysis Analysis: the two hosts share one source text; Cursor's copies appear to be the newer revision. They add features (inference hooks, MCP allowlisting, Team Bot management) and rename "plugins" to "connected apps" and "Team Setup" to "Team Setup Scripts". Where they conflict on whether a capability exists (DLP/inference hooks), the Cursor page is the better guide to current behaviour. Still, cite both and say the xAI copy may lag. The value-level details (domains, installer URLs, file paths, commands) agree across hosts, apart from the App Store link form, so either can be quoted for those. Analysis: most failed rollouts trace to three items the docs repeatedly flag: the missing nested `*.*.cursorvm.com` wildcard, Privacy Mode (Legacy), and device-trust rules (FastPass/compliant device) that the Linux hosted computer cannot satisfy. Front-loading steps 1–3 removes them. Analysis: Teams (self-serve) admins can do only part of this list. Network Controls, Team Setup/Secrets, Enforce Auto-review, Auto-review rules, Action Recording, audit logs and SCIM are Enterprise only, so a security-sensitive organization effectively needs Enterprise. ## Recommendations (These are the wiki's recommendations, derived from the sources.) - **Use Teams** if you only need access for everyone and accept per-member settings. **Use Enterprise** if you need destination allowlisting, enforced approvals, fleet tooling or audit export. - Run steps 1–3 before enabling. A member who signs in before the gateway is fixed sees setup hangs that look like product bugs. - If you pick an allowlist network mode, also turn on **Apply to MCP servers** and add connector hosts first, or connectors bypass the list (Cursor docs). - Pin and mirror installers, but budget for forced updates: support limits are unpublished. - Treat disabling as non-revocation. Offboarding = remove from team/group + revoke IdP sessions. ## Pages Compared [[concepts/teams-and-enterprises]], [[concepts/enterprise-deployment]], [[concepts/identity-and-access]], [[concepts/private-networks]], [[concepts/proxies]], [[syntheses/security-model]] ## Sources - raw/web_community-teams-and-enterprises-md.md — https://docs.x.ai/grok-bot/teams-and-enterprises.md - raw/web_community-teams-md.md — https://cursor.com/docs/grok-bot/teams.md - raw/web_community-deployment-md-2.md — https://docs.x.ai/grok-bot/deployment.md - raw/web_community-deployment-md.md — https://cursor.com/docs/grok-bot/deployment.md - raw/web_community-identity-and-access-md.md — https://docs.x.ai/grok-bot/identity-and-access.md - raw/web_community-identity-md.md — https://cursor.com/docs/grok-bot/identity.md - raw/web_community-private-networks-md-2.md — https://docs.x.ai/grok-bot/private-networks.md - raw/web_community-private-networks-md.md — https://cursor.com/docs/grok-bot/private-networks.md - raw/web_community-proxies-md-2.md — https://docs.x.ai/grok-bot/proxies.md - raw/web_community-proxies-md.md — https://cursor.com/docs/grok-bot/proxies.md - raw/web_community-security-md.md — https://cursor.com/docs/grok-bot/security.md - raw/web_community-security-md-2.md — https://docs.x.ai/grok-bot/security.md --- title: "Getting Started with Grok Bot" type: synthesis tags: [grok-bot, setup, install, onboarding, first-task] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-get-started-md-2.md", "raw/web_community-get-started-md.md", "raw/web_community-mobile-md.md", "raw/web_community-faq-md.md", "raw/web_community-approvals-security-and-privacy-md.md", "raw/web_community-grok-bot-md.md", "raw/web_community-overview-md.md", "raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md"] --- # Getting Started with Grok Bot This page merges the two official get-started guides, docs.x.ai/grok-bot/get-started.md and cursor.com/docs/grok-bot/get-started.md, into one first-run sequence. Where the two guides differ, both versions are shown. ## Comparison ### Prerequisites - **An eligible plan.** - The x.ai guide lists every paid individual Cursor plan, the Cursor Teams plan, or a linked individual SuperGrok, SuperGrok Plus or SuperGrok Heavy subscription. - The Cursor guide adds "a seat on a self-serve Grok Business plan". - See [[concepts/plans-and-access]]. - **Cloud data storage.** - x.ai wording: accounts using Legacy Privacy Mode "must move to a supported Cursor data setting before Grok Bot can start" (Cursor privacy settings). - Cursor wording: accounts on Privacy Mode (Legacy) "are prompted to switch to Privacy Mode". - **A first job.** You need an app or website where the first Bot can do useful work. ### Step-by-step (desktop) | Step | x.ai guide | Cursor guide | |---|---|---| | 1. Install | Download page [x.ai/bot](https://x.ai/bot). On macOS, choose Apple silicon or Intel, open the disk image, drag **Grok Bot** to **Applications**, and choose **Open** if asked. On Windows, choose x64 or Arm64 and run the installer. On Linux, use **More downloads** on x.ai/bot for a `.deb`, an `.rpm` or an AppImage | **Download Grok Bot** setup row on the Grok Bot page of the Cursor dashboard (cursor.com/dashboard/bot). Without dashboard access, ask your admin for the link | | 2. Sign in | **Sign in** on the welcome screen (or **Sign In with Cursor** from Settings), then finish in the browser. If your access is from SuperGrok, use **Settings → Usage & Billing** → **Get Access with Grok** on desktop | **Sign in**, then finish in the browser. "There is no separate Grok Bot login". Existing Cursor SSO configuration applies | | 3. First Bot | A first-run tour asks which tools you use. These answers only shape suggestions and connect nothing. The computer is set up in the background, and **Meet a future teammate** offers suggested Bots or **Create your own** (name, one primary job, description). Add more with **New → Create new Bot** | After the intro, the app creates a first Bot named **Grok Bot** and opens a chat. Add more with **New Bot** (the **+** button). Teams can use the command palette → **Create new Team Bot** | | 4. First task | Type, or dictate with **Start voice input** (`Cmd/Ctrl+D`). You can also use **Start voice chat** | Type the request. The guide gives the same five-part request formula | | 5. Tool sign-in | Choose **Take over**, enter the password, passkey, 2FA code or CAPTCHA yourself, then choose **I'm done**. Alternatively, install a plugin from **Connect apps** | Click **Open computer**, take over, enter the credentials yourself and return control. "The Bot doesn't see your password." Alternatively, connect an app | | 6. Review | Ask for revisions, name lasting preferences, then save the process as a skill or routine | Same | Architecture checks from the x.ai guide: - **macOS:** **Apple menu → About This Mac**. A **Chip** field means Apple silicon, and **Processor** means Intel. - **Windows:** **Settings → System → About → System type**. - **Linux:** `uname -m`. `x86_64` means x64, and `aarch64` means Arm64. - **Updates:** they are automatic, or use **Check for Updates** in **Settings → Updates**. ### A strong first request (both guides) State these five things: 1. **Outcome** 2. **Sources** 3. **Constraints** 4. **Deliverable** 5. **Review point**: where the Bot should stop for you For a five-minute result that needs no login, both guides suggest attaching a document and asking for a five-bullet summary that lists every date, decision and open question with citations, and that tells the Bot not to change the source file. The x.ai guide's second example compares activation in an analytics dashboard and adds "Do not change any dashboards. Ask me to sign in if needed." ### First approval prompts Before a Bot changes external systems, both guides send you to the approvals docs: x.ai's "Approvals, security, and privacy" and Cursor's "Approvals and Auto Review". When an action needs approval, the conversation shows the proposed operation and its inputs: - **Allow once** continues that action. - **Always allow** can save a matching rule. - **Deny** blocks the action. The controls are the same on iPhone. An approval from work you didn't start (a routine, a trigger, or another Bot) expires after about 10 minutes and shows **Expired**. Details are in [[concepts/approvals]]. ### iOS / Android first run - **Requirements:** iOS 18 or later (iPadOS 18 for iPad) or Android 9 or later. - **Sign-in:** **Log In or Sign Up**, then Cursor authentication in the browser. - **SuperGrok users:** **Link Grok Account**, then **Finished Linking? Refresh My Status**. - **New users** get a tour, choose a first Bot and wait for the computer to be set up. **Existing users** go straight to their synced Bot list. See [[concepts/mobile-app]]. ## Analysis - **Platform availability depends on the source and its date.** - x.ai get-started: desktop on macOS, Windows or Linux, plus a mobile page. - Cursor get-started: macOS, Windows and Linux desktop plus iPhone, and it mentions no Android. - x.ai mobile and FAQ pages: iPhone, iPad and Android 9 or later. - Cursor overview: "macOS, Windows, Linux, and iOS". - Gigazine (2026-08-12) reported that, at launch, the official docs listed Windows and macOS desktop plus iOS, with no Android version and no official Linux desktop app at the time. - All official pages were fetched 2026-10-21. The launch-time and current docs disagree, and Cursor's overview and get-started pages mention only iPhone, while Cursor's own deployment page lists Android 9+ as well; the hosts don't disagree, Cursor's intro pages just omit it. - **The first-Bot flow differs between hosts.** The x.ai guide describes a suggestion screen (**Meet a future teammate**). The Cursor guide describes an auto-created Bot named **Grok Bot**. The wiki can't tell which is current, so expect either. - **The labels for the same takeover action differ** (**Take over**/**I'm done** vs **Open computer**). Analysis: the button names may vary by host or app version. ## Recommendations - Start with the attach-a-document task. It proves the loop works without any connector or website login. - Put "do not change X / ask me before Y" in the first task, and in the Bot's description for standing limits. - If sign-in fails or the computer never finishes setting up, see [[syntheses/troubleshooting]]. The Cursor guide says most setup problems clear after you fully quit and reopen the app. ## Pages Compared [[syntheses/overview]], [[concepts/plans-and-access]], [[concepts/mobile-app]], [[concepts/bots]], [[concepts/approvals]], [[concepts/apps-and-connectors]], [[concepts/skills-routines-and-automations]], [[syntheses/troubleshooting]] ## Sources - raw/web_community-get-started-md-2.md — https://docs.x.ai/grok-bot/get-started.md - raw/web_community-get-started-md.md — https://cursor.com/docs/grok-bot/get-started.md - raw/web_community-mobile-md.md — https://docs.x.ai/grok-bot/mobile.md - raw/web_community-faq-md.md — https://docs.x.ai/grok-bot/faq.md - raw/web_community-approvals-security-and-privacy-md.md — https://docs.x.ai/grok-bot/approvals-security-and-privacy.md - raw/web_community-grok-bot-md.md — https://cursor.com/docs/grok-bot.md - raw/web_community-overview-md.md — https://docs.x.ai/grok-bot/overview.md - raw/web_community-spacexai-has-released-grok-bot-an-ai-agent-that-works-24-7-i.md — https://gigazine.net/gsc_news/en/20260812-spacexai-grok-bot/ (third-party) --- title: "Grok Bot Overview" type: synthesis tags: [grok-bot, overview, bots, cloud-computer, start-here] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-overview-md.md", "raw/web_community-grok-bot-md.md", "raw/web_community-faq-md.md", "raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md", "raw/web_community-deployment-md.md", "raw/web_community-get-started-md-2.md", "raw/web_community-official-grok-bot-breakdown-2026.md"] --- # Grok Bot Overview Grok Bot is a product for running **Bots**: named AI teammates with a job, a memory and their own persistent cloud computer (browser, filesystem and terminal). You give a Bot work by messaging it, and it carries multi-step tasks through in your real apps and websites. It is not a chat assistant that hands you drafts. The official docs live on two hosts, docs.x.ai/grok-bot/* and cursor.com/docs/grok-bot/*. The x.ai news post calls it "beta". For the product facts (makers, dates, platforms), see [[entities/grok-bot]]. ## Comparison ### Grok Bot vs a chat assistant | Dimension | Chat assistant | Grok Bot (per official docs) | |---|---|---| | Output | Answers and drafts | Finished work inside the target tools; drafts that you approve before they are sent | | Where it runs | Your chat session | A persistent cloud computer that belongs to your account. Work continues when your laptop or phone is closed | | Memory | Usually resets per chat | Each named Bot keeps its memory, files, browser sessions and preferences across sessions | | Parallelism | One conversation | Several Bots work in parallel, message each other, use group chats and hand off tasks | | Setup | Prompting | "Setup is a message, not a workflow builder". You create a Bot, describe its job and grant access when it asks | | Unsupported sites | N/A | Uses connectors where they exist and computer use for everything else. A site can still block automation, and the Bot then hands that step to you | The x.ai FAQ makes the same point: Bots use a cloud computer, connected tools, websites and files to *complete work*, they don't only answer questions. ### The parts | Part | What it is | Page | |---|---|---| | Bots | Named teammates with one primary job each. The docs say focused Bots build more useful context than one catch-all Bot | [[concepts/bots]] | | Team Bots | A Bot that is shared with a team so that teammates chat with the same Bot | [[concepts/team-bots]] | | Messaging | Text, dictation, voice chat, group chats, threads and handoffs | [[concepts/messaging-and-collaboration]] | | Cloud computer | One persistent computer per user, shared by all of that user's Bots. Each Bot gets its own screen, and one Bot runs one computer-use task at a time | [[concepts/cloud-computer]] | | Connectors / apps | Structured integrations. The x.ai docs call them "plugins" or "connectors", the Cursor docs call them "connected apps" | [[concepts/apps-and-connectors]] | | Skills & routines | A skill is how to do a task. A routine assigns a workflow to one Bot on a schedule or, where supported, after an event | [[concepts/skills-routines-and-automations]] | | Approvals | **Allow once** / **Always allow** / **Deny**, Auto Review rules, and takeover for passwords and CAPTCHAs | [[concepts/approvals]] | | Tag @bot on X | Start work from an X post | [[concepts/tag-bot-on-x]] | | Enterprise controls | Admin controls, identity, private networks, proxies and inference hooks | [[concepts/teams-and-enterprises]] | ### Start here: "I want to…" | I want to… | Go to | |---|---| | Install and run a first task | [[syntheses/getting-started]] | | Know whether my plan includes it | [[concepts/plans-and-access]] | | Use it on a phone | [[concepts/mobile-app]] | | Create, edit, share or delete Bots | [[concepts/bots]] | | Share one Bot with my team | [[concepts/team-bots]] | | Understand the shared computer and its limits | [[concepts/cloud-computer]] | | Connect Gmail, Slack, a CRM and so on | [[concepts/apps-and-connectors]] | | Automate a recurring job | [[concepts/skills-routines-and-automations]] | | Control what a Bot may do without asking | [[concepts/approvals]], [[syntheses/security-model]] | | Attach files or review outputs | [[concepts/files-and-results]] | | Change settings or notifications | [[concepts/settings-and-notifications]] | | Find role ideas and starter prompts | [[concepts/use-cases]] | | Understand data handling | [[concepts/privacy-and-data]] | | Roll it out to an organization | [[syntheses/enterprise-rollout-guide]], [[concepts/enterprise-deployment]], [[concepts/identity-and-access]] | | Put Bots behind our network or proxy | [[concepts/private-networks]], [[concepts/proxies]] | | Inspect model requests with our security tooling | [[concepts/inference-hooks]] | | Fix something broken | [[syntheses/troubleshooting]] | | Find which doc covers what | [[summaries/docs-map]] | | See what shipped and when | [[summaries/release-history]] | ## Analysis - **Who ships it.** The cloud computer runs in "Cursor's cloud" (both overview pages, and the Cursor deployment page). Grok Bot also uses your Cursor account for sign-in. xAI's site hosts the download page (x.ai/bot) and the launch news. In practice (the wiki's inference) Cursor account sign-in and data/privacy settings apply, including for access that comes from a linked SuperGrok subscription. - **The shared computer is the key mental model.** All of a user's Bots share one computer, including its files, browser sessions and app logins. The x.ai FAQ says outright: "Do not use separate Bots as a security boundary." Between users, the docs describe isolation as strict. - **The two hosts are mostly the same text with drift.** The two overview pages differ on platforms and plans: - The x.ai overview lists iOS and Android, and says the mobile app runs on iPhone, iPad and Android. - The Cursor overview (cursor.com/docs/grok-bot.md) lists only iOS, and says the mobile app runs on iPhone. - Only the Cursor overview adds a "self-serve Grok Business plan" seat as a way to get access. - Both pages were fetched 2026-10-21. See [[summaries/docs-map]] for the full divergence list. - **Third-party framing (attributed).** Vellum's breakdown calls it "xAI's early-beta AI agent", says it is a separate product from consumer Grok, the xAI API and Grok Build, and lists limits such as no self-hosting and no model choice. Those claims are Vellum's and are not stated in the official docs. ## Recommendations - Use Grok Bot if the work spans several tools, has a clear finish line, and benefits from running while you're away. The official "good first handoff" example is a CRM-plus-research outreach task that leaves drafts for you to approve. - Give each Bot one job. Put standing limits in the Bot's description, and add narrow **Ask first** rules for sending, publishing, deleting, purchasing or production changes (x.ai FAQ). - Don't use separate Bots to isolate sensitive access. Anything on the computer is available to every Bot you run. - For consequential decisions, ask the Bot to re-check the current source instead of relying on memory (both overviews). ## Pages Compared [[entities/grok-bot]], [[syntheses/getting-started]], [[concepts/plans-and-access]], [[concepts/cloud-computer]], [[concepts/bots]], [[concepts/approvals]], [[concepts/teams-and-enterprises]], [[summaries/docs-map]] ## Sources - raw/web_community-overview-md.md — https://docs.x.ai/grok-bot/overview.md - raw/web_community-grok-bot-md.md — https://cursor.com/docs/grok-bot.md - raw/web_community-faq-md.md — https://docs.x.ai/grok-bot/faq.md - raw/web_community-grok-bot-is-now-included-with-more-plans-spacexsi.md — https://x.ai/news/grok-bot-more-plans - raw/web_community-deployment-md.md — https://cursor.com/docs/grok-bot/deployment.md - raw/web_community-get-started-md-2.md — https://docs.x.ai/grok-bot/get-started.md - raw/web_community-official-grok-bot-breakdown-2026.md — https://www.vellum.ai/blog/official-grok-bot-breakdown (third-party) --- title: "Grok Bot Security Model" type: synthesis tags: [security, threat-model, controls, network-policy, auto-review, isolation, enterprise] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-security-md.md", "raw/web_community-security-md-2.md", "raw/web_community-security-faq-md.md", "raw/web_community-security-faq-md-2.md", "raw/web_community-approvals-security-and-privacy-md.md", "raw/web_community-teams-md.md", "raw/web_community-teams-and-enterprises-md.md", "raw/web_community-identity-md.md", "raw/web_community-private-networks-md.md", "raw/web_community-inference-hooks-md.md", "raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md", "raw/web_community-official-grok-bot-breakdown-2026.md"] --- # Grok Bot Security Model ## Comparison **Architecture (official).** Desktop and mobile apps are thin clients. They send each request to Cursor's servers. Work then runs on the member's hosted computer (a per-user Firecracker microVM), prompts go to model providers, and tool calls reach third-party services through connected apps. A control sits on each path. **Threats the docs address.** - **Over-broad access.** Any Bot can use whatever the member's computer is signed into. - **Prompt injection** from web pages, app results, and command output. - **Data egress** to arbitrary destinations. - **Consequential actions** such as sending, purchasing, deleting, and production changes. - **Credential exposure** in chat or scripts. - **Leftover access** after a project ends. | Control | What it protects | Where configured | Plan / default | |---|---|---|---| | Per-user isolation | One user's computer cannot reach another's | Built in (Firecracker microVM) | All | | Bots act as the member | No machine identity. A Bot never exceeds the member's access. Exception: team-managed connectors may use team or service-account credentials | Built in | All | | Connector token custody | OAuth tokens stay on Cursor's backend, never on the computer | Built in | All | | Take over / secure secret request | Passwords, 2FA, and payment details stay away from the model and transcript | In chat | All | | Approvals (**Allow once** / **Always allow** / **Deny**) | Consequential actions | Approval card | All | | Auto Review + personal rules | Pre-run review of shell, app calls, computer use, automation writes, delegation | Settings → General → Auto-review | All. **Ask first** wins | | **Enforce Auto-review** + team rules | Members can't turn review off. Team rules are locked | Dashboard Grok Bot page | Enterprise. **Off by default** | | Connector policy | Which MCP servers or plugins are usable | Dashboard **Plugins & MCPs** | All plans. MCP allowlist is Enterprise | | **Network Controls** | Which destinations computers reach | Dashboard → Network / **Grok Bot Network Access** | Enterprise. **No policy = allow-all** | | **Apply to MCP servers** | Holds remote MCP servers to the allowlist | Same panel, allowlist modes only | Enterprise. **Off by default** (cursor.com only) | | Local execution policy | Commands and files on the member's own machine | Settings (member) and team ceiling | Member default **Ask every time**. Team default **Always allow**, which leaves the choice to members | | **Allow Sudo on Bot Computers** | Root on the cloud computer | Dashboard → Bot Computers | Enterprise. **On by default** | | **Allow Local Egress Routing** | Members routing computer traffic through their desktop | Dashboard → Bot Computers | Enterprise. **On by default** | | Team Rules | Required guidance for every Bot | Dashboard | Teams and Enterprise | | Public template sharing | Bot configs published outside the team | Dashboard | Enterprise starts **off**. Other teams start allowed | | Inference hooks | Your DLP or AI-security check on model inputs | Dashboard → Security & Compliance | Enterprise **beta**. Starts in Shadow mode | | Audit logs / Action Recording / OTel Export | Control-plane events and Bot actions (90-day internal store) | Dashboard / Team Settings | Enterprise. Action Recording **off by default** | | SSO (SAML 2.0) / SCIM | Who can sign in. Deprovisioning | Cursor SSO app in your IdP | SSO works with Okta, Entra, Google Workspace, and OneLogin. SCIM is Enterprise | | Computer terminate / recreate / delete | Ends current work. Wipes data | **Manage Member Computers and Data** | Enterprise | ## Analysis **Hard boundaries vs judgment layers.** The docs separate controls that "don't depend on a model's judgment" (per-action approvals, the network policy, per-user isolation) from Auto Review, which is model-based. Both hosts say defenses against prompt injection reduce but do not eliminate the risk (x.ai: "reduce, but do not eliminate"; Cursor: "reduce, but don't eliminate"). Outside content is marked as untrusted data. **The shared-computer boundary is the key weakness to design around.** Within one user, Bots isolate personalities and workspaces, not compute. In practice, a low-trust Bot (for example, one that reads arbitrary web pages) shares cookies and credentials with a high-trust one. The documented mitigation is a separate Cursor user per credential set. **Two egress paths, two controls.** Blocking a connector doesn't block the service's website. Allowlisting destinations doesn't restrict remote MCP servers unless **Apply to MCP servers** is on. Self-serve Teams cannot set a destination allowlist at all. **Defaults lean permissive.** The out-of-the-box state is: - network allow-all - sudo on - local egress routing on - the team local-execution ceiling at **Always allow** (members still default to asking) - Auto-review enforcement off - Action Recording off - Cloud Agent delegation on (a Teams-level control) Enterprise admins must turn the restrictive settings on deliberately. **Host divergences that matter for a review (as of fetch 2026-10-21):** - **Model allowlist.** cursor.com says it "does not govern Grok Bot". docs.x.ai says it is honored by default but enforcement "is not guaranteed". - **DLP.** docs.x.ai says "Dedicated data loss prevention hooks are not available". cursor.com documents inference hooks as an Enterprise beta. - **Personal Auto-review rules.** docs.x.ai says they are stored per desktop. cursor.com says they are saved to the account. - **Conversation content export.** It is opt-in and off by default, and documented only on docs.x.ai. **Third-party analyses (medium confidence, attributed).** - Davies Meyer frames an agent with real logins as "an identity topic". It recommends dedicated service accounts per bot, broad-read/narrow-write permissions, logging every write action, a GDPR review, and defenses against indirect prompt injection. - Vellum (a competing assistant vendor, so a commercial source) notes that Bot boundaries "aren't security boundaries", that there is no self-hosting, and that privacy posture depends on the access route. Both points about shared computers and no self-hosting match the official docs. Davies Meyer's "service accounts per bot" advice goes beyond the official model: officially, Bots act as the member, and separation comes from a separate Cursor user. ## Recommendations Drawn from the Recommended configuration in cursor `teams.md`, reordered and extended by the wiki (items 6 and 7 are the wiki's additions; item 10 comes from the Identity section of `security.md`): 1. **Set Network Controls** to **Defaults + Team Allowlist** or **Team Allowlist Only**. Add connector hosts, then turn on **Apply to MCP servers**. Note: in an allowlist mode, web search is off and web fetch follows the list (docs.x.ai security-faq). 2. **Audit the connector policy** before enabling. Every permitted connector reaches every Bot. 3. Set the team **Execution on Local Computer** ceiling to **Never allow** unless Bots need member machines. 4. Turn on **Enforce Auto-review** and add narrow team **Ask first** rules for sending, publishing, purchasing, deletion, and production changes. 5. Keep **public template sharing** off. Consider turning off **Allow Sudo on Bot Computers** and **Allow Local Egress Routing** if they aren't needed. This is the wiki's analysis; the docs only state the defaults. 6. Turn on **Action Recording** plus OTel Export for Bot-action visibility. Pilot **inference hooks** in Shadow mode. 7. Put credentials in **Team Secrets**, never in setup scripts. Use a separate Cursor user per credential set. 8. **Disable Cloud Agent spawning** if you don't need delegation. 9. **Gate sign-in to managed devices** through your identity provider. 10. For offboarding, remove the member from the team or turn off Grok Bot for their group, *and* revoke their IdP sessions. Terminating a computer alone doesn't remove access. ## Pages Compared [[concepts/approvals]], [[concepts/privacy-and-data]], [[concepts/inference-hooks]], [[concepts/identity-and-access]], [[concepts/private-networks]], [[concepts/proxies]], [[concepts/cloud-computer]], [[concepts/apps-and-connectors]], [[concepts/teams-and-enterprises]], [[syntheses/enterprise-rollout-guide]] ## Sources - raw/web_community-security-md.md (https://cursor.com/docs/grok-bot/security.md) - raw/web_community-security-md-2.md (https://docs.x.ai/grok-bot/security.md) - raw/web_community-security-faq-md.md (https://cursor.com/docs/grok-bot/security-faq.md) - raw/web_community-security-faq-md-2.md (https://docs.x.ai/grok-bot/security-faq.md) - raw/web_community-approvals-security-and-privacy-md.md (https://docs.x.ai/grok-bot/approvals-security-and-privacy.md) - raw/web_community-teams-md.md (https://cursor.com/docs/grok-bot/teams.md) - raw/web_community-teams-and-enterprises-md.md (https://docs.x.ai/grok-bot/teams-and-enterprises.md) - raw/web_community-identity-md.md (https://cursor.com/docs/grok-bot/identity.md) - raw/web_community-private-networks-md.md (https://cursor.com/docs/grok-bot/private-networks.md) - raw/web_community-inference-hooks-md.md (https://cursor.com/docs/grok-bot/inference-hooks.md) - raw/web_community-grok-bot-xai-s-ai-teammates-with-their-own-computer.md (https://ai-solutions.daviesmeyer.com/en/blog/grok-bot-ki-teammates-xai), third-party - raw/web_community-official-grok-bot-breakdown-2026.md (https://www.vellum.ai/blog/official-grok-bot-breakdown), third-party --- title: "Troubleshooting Grok Bot" type: synthesis tags: [troubleshooting, faq, recovery, sign-in, proxies, routines, approvals] created: 2026-10-21 updated: 2026-10-21 confidence: medium sources: ["raw/web_community-troubleshooting-md.md", "raw/web_community-faq-md.md", "raw/web_community-work-md.md", "raw/web_community-computer-and-apps-md.md", "raw/web_community-settings-md.md", "raw/web_community-proxies-md.md", "raw/web_community-security-faq-md.md", "raw/web_community-security-md.md", "raw/web_community-teams-md.md", "raw/web_community-identity-md.md", "raw/web_community-inference-hooks-md.md", "raw/web_community-tag-on-x-md.md", "raw/web_community-get-started-md.md", "raw/web_community-deployment-md.md", "raw/web_community-computers-md-2.md", "raw/web_community-private-networks-md.md", "raw/web_community-approvals-security-and-privacy-md.md"] --- # Troubleshooting Grok Bot ## Comparison Symptom → cause → fix. Start with the least destructive step. Cloud work can continue while the app is disconnected. | Symptom | Likely cause | Fix | |---|---|---| | Sign-in doesn't complete | The browser auth handoff didn't return | Keep Grok Bot open during browser auth. Confirm the Cursor sign-in succeeded. Return to the app manually. Choose **Sign in** again, or **Sign In with Cursor** in Settings. Confirm the account has access. With SSO, use the org login, not a personal account | | Error mentioning Legacy Privacy Mode | Privacy Mode (Legacy) forbids the storage Grok Bot needs | Change the Cursor account data setting or ask your admin ([[concepts/privacy-and-data]]) | | **User is not assigned to this application** | The Cursor SSO app is assigned only to some groups | Widen assignment on the *existing* Cursor app (and SCIM app). Don't create a second app ([[concepts/identity-and-access]]) | | Stuck on **Starting your computer** / **Updating your computer** | Initial setup or an image update can take several minutes | Wait while progress changes. If it stalls: Retry, restart Grok Bot, check for an app update, then **Settings → Computer → Update Bot's Computer** | | Computer setup hangs on a corporate network, works on a hotspot | A TLS-inspecting gateway (e.g. Zscaler) blocks or inspects `cursorvm.com` | Allow `*.cursorvm.com` **and** `*.*.cursorvm.com`, exempt them from inspection and buffering, and apply to off-network profiles ([[concepts/proxies]]) | | Works in the office, fails at home | Exceptions weren't applied to the gateway's off-network profile | Apply them to every location and profile | | Computer unreachable | Transient, or the computer is broken | 1) **Retry** or reopen. 2) Restart the app. 3) **Recover computer**. 4) **Update Bot's Computer**. On mobile: **Settings → Bot → Bot Computer → Reset Computer**. 5) Wait for the replacement. 6) **Reset** only if you accept losing recent unsynced work | | Bot appears stuck | Waiting on a page, question, approval, login, CAPTCHA, or secret request, or another computer-use task is active on its screen | Check the status. Open the computer. Answer what's pending. Send a short redirect or "Stop now" | | Usage exhausted / spend limit reached | Weekly usage used up or the on-demand limit hit | Review **Usage & Billing** or the account access page ([[concepts/plans-and-access]]) | | Site keeps asking for login | Session expiry or per-action verification | Take over, sign in, complete 2FA or CAPTCHA, confirm the page loaded, tell the Bot to continue. This can't always be avoided | | Plugin won't install or authenticate | Not installed, auth revoked, or needs admin config | **Connect apps → Your plugins → Installed**. Re-run auth with the intended account. Check org variables. If revoked, remove and reconnect ([[concepts/apps-and-connectors]]) | | App shows **Disabled by team admin** | The team connector policy blocks it | Admin enables it in Teams Marketplace and, if the team uses an MCP allowlist, adds the server URL there. Member restarts the app | | MCP connector refused "not on your team's Grok Bot network allowlist" | **Apply to MCP servers** is on and a host is missing | Admin adds the host to **Allowed destinations** | | Attachment unreadable | Over 25 MB (200 MB video), more than 6 attachments on desktop, encrypted, still uploading, or an unsupported type | Fix the cause, or export as PDF, CSV, plain text, or an image | | Routine didn't run | Disabled, wrong schedule or time zone, owning Bot gone, plugin auth lapsed, source unreachable, or usage paused | Check each, then the run history. **Test run** performs real actions, so use safe inputs ([[concepts/skills-routines-and-automations]]) | | Action keeps requiring approval | A matching **Ask first** rule, possibly an admin team rule | Check **Settings → General → Bot → Auto-review** ([[concepts/approvals]]) | | Approval card not actionable / **Expired** | An unattended approval expired after ~10 min | Reject or cancel it. Ask the Bot to retry or regenerate with corrected scope | | Local computer work refused | Local execution policy, or the team ceiling | Review **Execution on Local Computer** / **Execution on this computer** | | @bot didn't reply on X | Region not supported, video or GIF in the post, @grok also tagged, moderation, account not linked, or usage used up | See [[concepts/tag-bot-on-x]] | | **Blocked by your organization's policy** | An inference hook denied the request | **Edit** or **Reset**. If blocked again after a reset, check the Bot's memory and files ([[concepts/inference-hooks]]) | | **Your organization's policy check is unavailable** | Block-mode hook failure or a tripped circuit breaker | Admin checks **Recent Errors** on the Inference Hooks page, or switches to **Allow the Request** to keep members working. After fixing a tripped circuit breaker: Test Connection, then **Reset** | | **Update required** covers the window | The build is below Cursor's minimum version, or past a maximum age while a newer release is available | Update the app (applies to held Linux packages too) | | Members must re-sign in to company tools | The computer was recreated, e.g. after an image update | Expected. Sign in again under IdP policy | | Some websites block Bots | They flag datacenter IPs. Egress uses shared static ranges | Get the ranges from your account team and allowlist them on your own services | | Network-client login lost (Tailscale etc.) | Recreated computer, session didn't survive | Authenticate again ([[concepts/private-networks]]) | ## Analysis **FAQ answers mapped to pages** (docs.x.ai `faq.md`): - Work continues with the laptop closed, Bots share one computer, and the parallel limit is one computer-use task per screen: [[concepts/cloud-computer]] - What needs approval: [[concepts/approvals]] - Data and privacy (no Legacy Privacy Mode; training follows Cursor settings): [[concepts/privacy-and-data]] - Cost, plans, and SuperGrok linking: [[concepts/plans-and-access]] - Supported platforms (macOS Apple silicon/Intel; Windows x64/Arm64; Linux x64/Arm64 as `.deb`, `.rpm`, or AppImage; iOS 18+; Android 9+) and dictation `Cmd/Ctrl+D`: [[concepts/mobile-app]], [[concepts/settings-and-notifications]] - Skill vs routine, and Teach a task (10-minute limit, gradual rollout): [[concepts/skills-routines-and-automations]] - Sharing a Bot and what deleting one does: [[concepts/bots]] Analysis: most "stuck" symptoms have one of two root causes. Either the Bot is waiting on a human step (approval, login, secret), or the device-to-computer link is broken by a network gateway. The rarer root cause is a broken computer. Check the conversation first, then the network, and only then use recovery actions. **Divergences.** docs.x.ai says **Reset** "restores the last saved snapshot". cursor.com says Reset "rebuilds the computer from your last saved data, the same as Recover". Both warn that recent changes can be lost. Cursor's `get-started.md` notes most setup problems clear after fully quitting and reopening the app. ## Recommendations 1. Try the reversible steps first: retry, reopen the conversation, restart the app, check for an app update (**Settings → Updates → Check for Updates → Restart to Update**). Updating the app does not reset the cloud computer. 2. Before you recover, copy important results into `/workspace` or attach them to the conversation (docs.x.ai `computer-and-apps.md`). 3. On corporate networks, run the proxy checks from the member's device. `curl -v https://api2.cursor.sh` should show the issuer **Amazon RSA**, and `nslookup test.us9.cursorvm.com` should resolve. 4. Before contacting support, collect: Grok Bot version, OS and version, the exact error, the Bot or routine name, the time and time zone, the full request or conversation ID (**Copy request ID**), and whether retry, restart, or update changed anything. Never include passwords, codes, keys, or secrets. ## Pages Compared [[concepts/cloud-computer]], [[concepts/apps-and-connectors]], [[concepts/approvals]], [[concepts/proxies]], [[concepts/identity-and-access]], [[concepts/inference-hooks]], [[concepts/tag-bot-on-x]], [[concepts/skills-routines-and-automations]], [[syntheses/getting-started]] ## Sources - raw/web_community-troubleshooting-md.md (https://docs.x.ai/grok-bot/troubleshooting.md) - raw/web_community-faq-md.md (https://docs.x.ai/grok-bot/faq.md) - raw/web_community-work-md.md (https://cursor.com/docs/grok-bot/work.md) - raw/web_community-computer-and-apps-md.md (https://docs.x.ai/grok-bot/computer-and-apps.md) - raw/web_community-settings-md.md (https://cursor.com/docs/grok-bot/settings.md) - raw/web_community-proxies-md.md (https://cursor.com/docs/grok-bot/proxies.md) - raw/web_community-security-faq-md.md (https://cursor.com/docs/grok-bot/security-faq.md) - raw/web_community-security-md.md (https://cursor.com/docs/grok-bot/security.md) - raw/web_community-teams-md.md (https://cursor.com/docs/grok-bot/teams.md) - raw/web_community-identity-md.md (https://cursor.com/docs/grok-bot/identity.md) - raw/web_community-inference-hooks-md.md (https://cursor.com/docs/grok-bot/inference-hooks.md) - raw/web_community-tag-on-x-md.md (https://docs.x.ai/grok-bot/tag-on-x.md) - raw/web_community-get-started-md.md (https://cursor.com/docs/grok-bot/get-started.md) - raw/web_community-deployment-md.md (https://cursor.com/docs/grok-bot/deployment.md) - raw/web_community-computers-md-2.md (https://docs.x.ai/grok-bot/computers.md) - raw/web_community-private-networks-md.md (https://cursor.com/docs/grok-bot/private-networks.md) - raw/web_community-approvals-security-and-privacy-md.md (https://docs.x.ai/grok-bot/approvals-security-and-privacy.md)