---
title: "Google AI Studio (Gemini) Provider"
type: entity
tags: [provider, model-switching, well-established, intermediate]
created: 2026-06-10
updated: 2026-06-10
sources: ["raw/docs-integrations-providers.md", "raw/docs-reference-environment-variables.md", "raw/02-install-and-setup.md", "raw/release-v0.8.0.md", "raw/release-v0.11.0.md", "raw/release-v0.12.0.md", "raw/release-v0.13.0.md"]
confidence: high
hermes_version: "v0.15.0"
---

## Overview

The **Google AI Studio provider** gives Hermes direct access to Gemini models through Google's AI Studio API — no OpenRouter proxy. Provider id: `gemini`. The native provider shipped in [[entities/version-v0.8.0]] with automatic models.dev registry integration; [[entities/version-v0.11.0]] routed Gemini through the native AI Studio API and added a separate **Google Gemini CLI OAuth** inference provider plus a Gemini TTS provider. Gemini models matter beyond the main slot: `google/gemini-3-flash-preview` is the default context-compression summary model, and Gemini Flash via OpenRouter is the default auxiliary model for vision and web summarization.

## Characteristics

- **Setup:** `GOOGLE_API_KEY` in `~/.hermes/.env` — get one at aistudio.google.com/app/apikey. `GEMINI_API_KEY` is an alias; `GEMINI_BASE_URL` overrides the endpoint.
- **Provider id:** `gemini` (one of the 28 `plugins/model-providers/` plugins as of [[entities/version-v0.15.0]])
- **Free-tier guard:** Gemini free-tier keys are **blocked at setup** with 429 guidance surfacing ([[entities/version-v0.12.0]]) — free keys rate-limit too aggressively for agent use; use a paid/billing-enabled key
- **Multimodal:** `video_analyze` — native video understanding — runs on Gemini and compatible multimodal models ([[entities/version-v0.13.0]]); `vision_analyze` returns raw pixels to Gemini as a vision-capable model (v0.14.0)
- **OAuth alternative:** Google Gemini CLI OAuth inference provider (added v0.11.0) for subscription-style auth instead of an API key
- **Auxiliary role:** default compression summary model is `google/gemini-3-flash-preview`; the setup wizard's Context Compression defaults to it

## How to Use

```bash
# Set the key (auto-saves to ~/.hermes/.env)
hermes config set GOOGLE_API_KEY your-ai-studio-key

# Then pick Gemini in the interactive picker
hermes model
```

Permanent config in `~/.hermes/config.yaml` (same pattern as other providers):

```yaml
model:
  provider: "gemini"
  default: "gemini-3-flash-preview"
```

Use Gemini Flash as the dedicated compression model while a different provider runs the main loop:

```yaml
compression:
  summary_provider: openrouter
  summary_model: google/gemini-3-flash-preview
```

Note: a model id like `google/gemini-3-flash-preview` is the OpenRouter-format slug; on the native `gemini` provider, use the bare model name.

## Related Entities

- [[entities/provider-openrouter]] — reaches Gemini models via the aggregator, and is the default path for Gemini-Flash auxiliary tasks
- [[entities/provider-anthropic]], [[entities/provider-openai-codex]] — sibling frontier providers
- [[entities/version-v0.8.0]] — native provider added; [[entities/version-v0.11.0]] — native AI Studio routing + Gemini CLI OAuth; [[entities/version-v0.12.0]] — free-tier key block; [[entities/version-v0.13.0]] — `video_analyze`
- [[concepts/model-switching]]
- [[concepts/auxiliary-models]] — where Gemini Flash does most of its work in practice