Gemini¶

The gemini component connects to the native Google Gemini REST API using streaming SSE.

Configuration¶

components:
  - name: gemini
    type: gemini
    metadata:
      default_model: gemini-2.0-flash
      api_key: AIza...     # or set GEMINI_API_KEY / GOOGLE_API_KEY env var
    defaults:
      temperature: 1.0
      max_tokens: 2048
      top_p: 0.9
      top_k: 40
      system: "You are a helpful assistant."

Metadata reference¶

Key	Required	Description
`default_model`	no	Model used when the request doesn't specify one. Defaults to `gemini-2.0-flash`.
`api_key`	no	Falls back to `GEMINI_API_KEY`, then `GOOGLE_API_KEY` environment variable.
`base_url`	no	API base URL. Defaults to `https://generativelanguage.googleapis.com`. Override for Vertex AI or proxies.

Supported parameters¶

Parameter	Supported	Notes
`temperature`	✓
`max_tokens`	✓	Maps to `maxOutputTokens`
`top_p`	✓
`top_k`	✓	Gemini-specific
`stop`	✓	Maps to `stopSequences`
`frequency_penalty`	—	Not supported; ignored
`presence_penalty`	—	Not supported; ignored
`seed`	—	Not supported; ignored
`system`	✓	Passed as `systemInstruction`
`tools`	✓	Mapped to `functionDeclarations`

Models¶

Any model available through the Gemini API can be specified in default_model or overridden per-request.

Model	Notes
`gemini-2.0-flash`	Fast, efficient — recommended default
`gemini-2.0-flash-lite`	Lightest and cheapest
`gemini-2.5-pro`	Most capable
`gemini-2.5-flash`	Balanced performance

reply = client.llm("gemini").chat("Hello", model="gemini-2.5-pro")

API key setup¶

Obtain a key from Google AI Studio:

export GEMINI_API_KEY=AIza...

The provider checks api_key metadata first, then GEMINI_API_KEY, then GOOGLE_API_KEY.

Multiple instances¶

components:
  - name: gemini-flash
    type: gemini
    metadata:
      default_model: gemini-2.0-flash

  - name: gemini-pro
    type: gemini
    metadata:
      default_model: gemini-2.5-pro
    defaults:
      max_tokens: 8192
      temperature: 0.7