noumen — coding agents are products. This is the library.

Setup

pnpm add @ai-sdk/google

import { AiSdkProvider } from "noumen";
import { createGoogleGenerativeAI } from "@ai-sdk/google";

const google = createGoogleGenerativeAI({
  apiKey: process.env.GEMINI_API_KEY!,
});

const provider = new AiSdkProvider({
  model: google("gemini-2.5-flash"),
  providerFamily: "google",
});

Options

All connection-level options come from createGoogleGenerativeAI.

Option	Source	Description
`apiKey`	`createGoogleGenerativeAI`	Google AI Studio API key.
`baseURL`	`createGoogleGenerativeAI`	Override the API base URL.
`headers`	`createGoogleGenerativeAI`	Custom headers sent with every request.

Streaming

AiSdkProvider routes through the AI SDK's doStream(). Gemini-specific mapping is handled by @ai-sdk/google:

Text parts become content deltas.
functionCall parts become tool_calls deltas (with JSON repair applied).
Usage metadata is emitted on finish.

Message conversion

Handled by the AI SDK's convertToModelMessages:

System messages become the systemInstruction config parameter.
OpenAI-style tool messages are converted to functionResponse parts.
Tool call ID to function name resolution is handled automatically.
Multiple tool results are batched into a single user turn.

Thinking budget

Gemini 2.5 supports thinkingConfig. Enable it via AgentOptions.thinking:

const agent = new Agent({
  provider,
  sandbox: LocalSandbox({ cwd: "." }),
  options: { thinking: { type: "enabled", budgetTokens: 8_000 } },
});

AiSdkProvider maps this to providerOptions.google.thinkingConfig = { thinkingBudget: 8000 }.

Gemini behind an OpenAI-compatible proxy

If you route Gemini traffic through an OpenAI-compatible gateway (e.g. your own metered proxy that exposes /v1/chat/completions), wrap it with @ai-sdk/openai and pin to .chat(...) — noumen will classify it as the openai family and map reasoning effort accordingly:

import { createOpenAI } from "@ai-sdk/openai";

const gateway = createOpenAI({
  baseURL: "https://my-proxy.example.com/google",
  apiKey: userJwt,
});
const provider = new AiSdkProvider({ model: gateway.chat("gemini-2.5-flash") });

Models

gemini-2.5-flash — fast, good default for coding tasks.
gemini-2.5-pro — higher capability for complex tasks.
gemini-2.0-flash — previous generation, still capable.

Google Gemini