Google Gemini

Setup

import { GeminiProvider } from "noumen";

const provider = new GeminiProvider({
  apiKey: "...",                // Google AI Studio API key
  model: "gemini-2.5-flash",   // default
});

Options

Streaming

The Gemini provider uses generateContentStream from the @google/genai SDK. It maps Gemini's response format to noumen's OpenAI-compatible chunks:

• Text parts become content deltas
• functionCall parts become tool_calls deltas
• usageMetadata on chunks is captured for token usage

Message conversion

Gemini has a unique message format. The provider handles the mapping:

• 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

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