@llm-optimize/gemini

Google Gemini adapter for @llm-optimize/core. Supports Gemini 1.5 Pro, Gemini 1.5 Flash, and Gemini 2.0 models.

Install

npm install @llm-optimize/gemini @llm-optimize/core

Usage

import { LLMOptimizer } from "@llm-optimize/core";
import { createGeminiProviderClient } from "@llm-optimize/gemini";

const ai = new LLMOptimizer({
  provider: "gemini",
  providerClientFactory: {
    create: () => createGeminiProviderClient({
      apiKey: process.env.GEMINI_API_KEY,
    }),
  },
});

const response = await ai.chat({
  model: "gemini-1.5-flash",
  messages: [
    { role: "system", content: "You are a concise assistant." },
    { role: "user", content: "What is the capital of France?" },
  ],
});

console.log(response.content);
console.log(response.usage); // { inputTokens, outputTokens, totalTokens }

System Messages

Gemini does not support a system role in the messages array. This adapter automatically extracts role: "system" messages and passes them as systemInstruction — you don't need to handle this yourself.

// This works correctly — system message is handled automatically
messages: [
  { role: "system", content: "Always respond in bullet points." },
  { role: "user", content: "List 3 benefits of TypeScript." },
]

Supported Models

| Model | ID | |---|---| | Gemini 1.5 Flash | gemini-1.5-flash | | Gemini 1.5 Pro | gemini-1.5-pro | | Gemini 2.0 Flash | gemini-2.0-flash | | Gemini 2.0 Flash Lite | gemini-2.0-flash-lite |

What's Included

• createGeminiProviderClient(options) — creates a fetch-based Gemini client
• GeminiProvider — class wrapper around any GeminiClient
• GeminiClient interface — implement this to bring your own HTTP client
• GeminiProviderOptions — { apiKey?, baseUrl? }