AI SDK Provider for GitHub Copilot

Vercel AI SDK community provider for GitHub Copilot — use streamText, generateText, and related AI SDK APIs with GitHub Copilot as the backend model.

---

Installation

1. Install and authenticate the Copilot CLI

Follow the Copilot CLI installation guide to install the CLI and ensure copilot is available in your PATH.

2. Add the provider

npm install @nomomon/ai-sdk-provider-github-copilot ai@^6.0.0

Quick Start

Non-streaming (generateText)

import { generateText } from "ai";
import { githubCopilot } from "@nomomon/ai-sdk-provider-github-copilot";

const { text } = await generateText({
  model: githubCopilot("gpt-5"),
  prompt: "Hello, Copilot!",
});

console.log(text);

Streaming (streamText)

import { streamText } from "ai";
import { githubCopilot } from "@nomomon/ai-sdk-provider-github-copilot";

const result = streamText({
  model: githubCopilot("gpt-5"),
  prompt: "Tell me a short story",
});

for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

Models

Use model IDs available via Copilot CLI. Run copilot -i /models to list available models in your environment.

Examples

Script	Description
npm run example:basic	Non-streaming generateText
npm run example:streaming	Streaming streamText
npm run example:tools	Model-level tools via Copilot defineTool
npm run example:tools-ai-sdk	Call-level AI SDK tool() with providerOptions bridge

Configuration

Provider settings

import { githubCopilot } from "@nomomon/ai-sdk-provider-github-copilot";

const model = githubCopilot("gpt-5", {
  model: "claude-sonnet-4.5", // Override model
  systemMessage: {
    content: "You are a helpful assistant specialized in code review.",
  },
  workingDirectory: "/path/to/project",
  provider: {
    type: "openai",
    baseUrl: "https://my-api.example.com/v1",
    apiKey: process.env.MY_API_KEY,
  },
});

Custom tools

Tools can be passed in two ways. When both are used, call-level tools are merged with model-level tools before creating the Copilot session.

1. Copilot's defineTool (model-level) — configure tools when creating the model:

import { defineTool } from "@github/copilot-sdk";
import { z } from "zod";
import { githubCopilot } from "@nomomon/ai-sdk-provider-github-copilot";

const model = githubCopilot("gpt-5", {
  tools: [
    defineTool("lookup_issue", {
      description: "Fetch issue details from our tracker",
      parameters: z.object({
        id: z.string().describe("Issue identifier"),
      }),
      handler: async ({ id }) => {
        return await fetchIssue(id);
      },
    }),
  ],
});

2. AI SDK tool() with providerOptions (call-level) — the AI SDK does not pass execute to providers. Use copilotToolOptions(execute) so the provider can convert the tool and use it as the Copilot handler:

import { copilotToolOptions, githubCopilot } from "@nomomon/ai-sdk-provider-github-copilot";
import { streamText, tool } from "ai";
import { z } from "zod";

const execute = async ({ city }: { city: string }) => ({
  city,
  temperature: `${20 + Math.floor(Math.random() * 15)}°C`,
  condition: "sunny",
});

const getWeather = tool({
  description: "Get the current weather for a city",
  inputSchema: z.object({
    city: z.string().describe("The city name"),
  }),
  execute,
  providerOptions: copilotToolOptions(execute),
});

const result = streamText({
  model: githubCopilot("gpt-5-mini"),
  tools: { get_weather: getWeather },
  prompt: "What's the weather in Tokyo?",
});

Tool support varies by model; verify with Copilot CLI or documentation.

Development

Tests live in tests/ (not colocated with source) to keep the published src/ tree clean and to exclude test files from the build. Each source module has a corresponding tests/*.test.ts file. Run npm test or npm run test:coverage for coverage with thresholds.

Limitations

• Requires Copilot CLI - Must be installed and authenticated
• Node.js >= 18 - Required runtime
• Image inputs - Copilot uses file path attachments; base64/data URLs may require temp files
• Unsupported parameters - temperature, maxTokens, topP, etc. are not supported by the Copilot CLI and will be ignored (warnings emitted)
• Structured outputs - Native JSON schema support may be limited; consider prompt engineering for structured responses
• Session-based - Each generate/stream creates a new session; no built-in multi-session continuity across separate AI SDK calls
• Model capabilities - Tools, reasoning, and other features may vary by model; capabilities are determined by Copilot CLI

Disclaimer

This is an unofficial community provider and is not affiliated with or endorsed by GitHub or Vercel. By using this provider:

• You understand that your data will be sent to GitHub's servers through the Copilot CLI
• You agree to comply with GitHub's Terms of Service
• You acknowledge this software is provided "as is" without warranties of any kind

License

MIT