Usage guide
Setup, environment variables, and Python/JavaScript/Go usage examples.
GoogleLLM
Gemini chat models from Google, called with your own Google API key or through Vertex AI. A good fit for fast, low-cost conversational agents with tool calling and optional thinking. Authenticates with GOOGLE_API_KEY or GOOGLE_APPLICATION_CREDENTIALS from your environment, or an explicitapi_key.
Fields
str
default:"UNSET"
Gemini model name, e.g. “gemini-3.5-flash-lite” or “gemini-2.5-flash”. Defaults to ‘gemini-3.5-flash-lite’.
float
default:"UNSET"
Sampling randomness, 0.0 to 2.0; higher is more varied. Ignored by Gemini 3.5 and newer models. Defaults to 0.7.
Literal['auto', 'required', 'none']
default:"UNSET"
Whether the model may call tools: “auto” lets it decide, “required” forces a tool call, “none” disables tools. Defaults to ‘auto’.
int | None
default:"8192"
Maximum number of tokens the model may generate in one response. Defaults to 8192.
float | None
default:"UNSET"
Nucleus sampling: only tokens within this cumulative probability, 0.0 to 1.0, are considered. Ignored by Gemini 3.5 and newer models. Defaults to None (not set).
int | None
default:"UNSET"
Only the K most likely tokens are considered at each step. Ignored by Gemini 3.5 and newer models. Defaults to None (not set).
float | None
default:"UNSET"
Penalizes tokens that have already appeared, nudging the model toward new topics. Positive values reduce repetition. Defaults to None (not set).
float | None
default:"UNSET"
Penalizes tokens in proportion to how often they have appeared. Positive values reduce repeated wording. Defaults to None (not set).
bool
default:"UNSET"
Set to True to call Gemini through Google Cloud Vertex AI instead of the public Gemini API. Defaults to False.
Mapping[str, Any] | None
default:"UNSET"
Vertex AI settings such as {“project”: “my-project”, “location”: “us-central1”}. Only used when vertexai is True. Defaults to None (not set).
int | None
default:"UNSET"
Token budget for internal reasoning on Gemini 2.5 models; 0 turns thinking off. Set to None to omit the thinking setting entirely. Defaults to 0.
int | None
default:"UNSET"
Random seed for more repeatable outputs across identical requests. Defaults to None (not set).
list | None
default:"UNSET"
List of Gemini safety settings (category and threshold) controlling which content is blocked. Defaults to None (not set).
Any | None
default:"UNSET"
Low-level HTTP options for the Google client, such as timeout, API version or base URL. Defaults to None (not set).
str | None
default:"UNSET"
Reasoning effort for Gemini 3 and newer models: “minimal”, “low”, “medium” or “high”. “minimal” acts as “low” on Gemini 3.7+. Defaults to None (not set).
bool | None
default:"UNSET"
Set to True to include the model’s thought summaries in the response. Defaults to None (not set).