Interface: GeminiGenerateContentAdapterOptions
Defined in: src/batteries/llm/gemini_generate_content/types.ts:343
Constructor and per-dispatch options.
Remarks
Layered exactly like the other batteries: constructor baseline, then executor-scope overrides, then per-iteration ctx.stash overrides, re-validated each iteration.
Properties
| Property | Type | Description | Defined in |
|---|---|---|---|
apiKey? | string | API key. Sent as x-goog-api-key unless useBearerAuth is set. | src/batteries/llm/gemini_generate_content/types.ts:347 |
autoAck? | boolean | Whether to acknowledge tool-call-free generations; defaults to true. | src/batteries/llm/gemini_generate_content/types.ts:404 |
baseURL? | string | Defaults to https://generativelanguage.googleapis.com/v1beta. | src/batteries/llm/gemini_generate_content/types.ts:357 |
bucketOrder? | ChatCompletionsBucketOrder | Order of context buckets in the system instruction. | src/batteries/llm/gemini_generate_content/types.ts:408 |
contextWindow? | number | Maximum token weight allowed for the assembled context. | src/batteries/llm/gemini_generate_content/types.ts:402 |
fetch? | { (input: URL | RequestInfo, init?: RequestInit): Promise<Response>; (input: string | URL | Request, init?: RequestInit): Promise<Response>; } | Injectable fetch, for tests or a custom transport. | src/batteries/llm/gemini_generate_content/types.ts:398 |
forgeToolsFilter? | (forged: ToolRegistry, ctx: DispatchContext) => ToolRegistry | Narrow the artifact-reader tools forged from prior-turn results before they merge in. | src/batteries/llm/gemini_generate_content/types.ts:428 |
helpers? | Partial<GeminiGenerateContentHelpers> | Override any translation helper; unset entries fall back to the default* implementation. | src/batteries/llm/gemini_generate_content/types.ts:426 |
localToolCallParser? | | ToolCallParserName | ToolCallParserFn | Recover tool calls the provider returned as prose rather than functionCall parts. | src/batteries/llm/gemini_generate_content/types.ts:414 |
maxOutputTokens? | number | Upper bound on generated tokens. | src/batteries/llm/gemini_generate_content/types.ts:361 |
model | string | Model id, e.g. gemini-2.5-flash-lite. Prefer a CONCRETE id over a floating alias. | src/batteries/llm/gemini_generate_content/types.ts:345 |
onPromptAssembled? | PromptAssembledObserverFn | Observe the assembled request the instant before dispatch. Remarks The honest way to check what a vendor actually received — see the validation battery's "Which API surface a rule applies to" guide. An ADK-control key: stripped from the wire body. | src/batteries/llm/gemini_generate_content/types.ts:424 |
onRawGeneration? | RawGenerationObserverFn | Observe the raw generation. Purely observational; errors swallowed. | src/batteries/llm/gemini_generate_content/types.ts:416 |
retry? | ChatCompletionsRetryConfig | Retry/backoff policy for transport failures. | src/batteries/llm/gemini_generate_content/types.ts:396 |
safetySettings? | GeminiSafetySetting[] | Per-category safety-filter overrides. | src/batteries/llm/gemini_generate_content/types.ts:373 |
spoolStore? | SpoolStore | Spool store backing artifact-handle results. | src/batteries/llm/gemini_generate_content/types.ts:406 |
stopSequences? | string[] | Sequences that halt generation. | src/batteries/llm/gemini_generate_content/types.ts:369 |
stream? | boolean | Use :streamGenerateContent (SSE) instead of :generateContent. | src/batteries/llm/gemini_generate_content/types.ts:359 |
temperature? | number | Sampling temperature. | src/batteries/llm/gemini_generate_content/types.ts:363 |
thinkingConfig? | GeminiThinkingConfig | Reasoning-budget controls. | src/batteries/llm/gemini_generate_content/types.ts:371 |
thoughtSignatureSentinel? | string | false | Sentinel to stamp on the first historical functionCall part that carries no thoughtSignature, or false to send history untouched. Remarks Gemini 3+ REJECTS a historical functionCall with no signature. When history did not originate from Gemini (a replay, a model switch, a cross-vendor panel) there is no genuine signature to send, and Google documents two portable sentinels for exactly that case — 'skip_thought_signature_validator' (Gemini API and Vertex) and 'context_engineering_is_the_way_to_go' (Gemini API only). Defaults to 'skip_thought_signature_validator' because the alternative is a hard 400 on any non-Gemini-originated tool history. It is nonetheless a PROVENANCE CLAIM about reasoning the model did not produce, and Google cautions it may degrade quality relative to a real signature — set false to opt out and surface the vendor's own error instead. | src/batteries/llm/gemini_generate_content/types.ts:392 |
thoughtSurfacing? | "all" | "all-self" | "latest-self" | Which reasoning to replay. | src/batteries/llm/gemini_generate_content/types.ts:410 |
timeoutMs? | number | Per-request timeout. Falls back to the dispatch abort signal when absent. | src/batteries/llm/gemini_generate_content/types.ts:394 |
tokenEncoding? | | TokenEncodingId | null | Encoding used for token accounting; null explicitly disables token accounting. | src/batteries/llm/gemini_generate_content/types.ts:400 |
toolConfig? | GeminiToolConfig | Force, permit, or forbid function calling. | src/batteries/llm/gemini_generate_content/types.ts:375 |
topK? | number | Top-k sampling cutoff. | src/batteries/llm/gemini_generate_content/types.ts:367 |
topP? | number | Nucleus-sampling threshold. | src/batteries/llm/gemini_generate_content/types.ts:365 |
unsupportedMediaPolicy? | UnsupportedMediaPolicy | What to do with a modality Gemini cannot accept. | src/batteries/llm/gemini_generate_content/types.ts:412 |
useBearerAuth? | boolean | Send the key as Authorization: Bearer instead of x-goog-api-key. Remarks Google's own endpoint wants x-goog-api-key; several gateways in front of it accept only Authorization. Default false. | src/batteries/llm/gemini_generate_content/types.ts:355 |