tool-search

---

copilot-api lets existing AI tools talk to a local endpoint while the gateway handles authentication, model discovery, protocol translation, streaming, tool calls, provider routing, and usage accounting.

It began as an independent Rust reimplementation of caozhiyuan/copilot-api and now evolves independently. See NOTICE.md for attribution.

[!IMPORTANT] This is unofficial community software. It is not affiliated with or endorsed by GitHub, Microsoft, OpenAI, or Anthropic. Review the disclaimer and use the project responsibly.

What you get

Capability	What it unlocks
OpenAI compatibility	Chat Completions, Responses, compaction, models, embeddings, images, and a local Files API.
Anthropic compatibility	Messages, token counting, streaming, thinking blocks, tools, prompt caching, images, PDFs, and local file references.
Coding-agent support	Audited integration paths for Claude Code and Codex CLI, including their current headers, request shapes, and SSE lifecycles.
Backend choice	Use GitHub Copilot, Codex OAuth, Anthropic, OpenAI-compatible APIs, Responses-compatible APIs, or your own provider aliases.
Local control	API keys, admin routes, model mappings, token budgets, rate limits, load shedding, and provider-only mode.
Operational visibility	Readiness and version endpoints, Prometheus metrics, structured logs, diagnostics, and a built-in usage dashboard.
Simple distribution	Install from crates.io, download a release binary, build from source, or run the published container image.

How it fits together

flowchart LR
    A[Claude Code] --> G[copilot-api<br/>localhost:4141]
    B[Codex CLI] --> G
    C[OpenAI / Anthropic SDKs] --> G
    D[Other compatible clients] --> G

    G --> E[GitHub Copilot]
    G --> F[Codex OAuth]
    G --> H[Third-party providers]

    G --> I[Local files]
    G --> J[Usage + metrics]

Clients keep using familiar APIs. The gateway resolves the requested model, selects the right upstream transport, translates the request and streaming response, and preserves the client-facing protocol.

Quick start

1. Install

With Cargo:

cargo install copilot-api --locked

Or download a prebuilt binary from the latest release. Release assets are published for Linux x86-64, macOS Apple Silicon, and Windows x86-64.

2. Start the gateway

copilot-api start

On first run, copilot-api starts GitHub's device-login flow, opens the authorization page when possible, stores the token in your local app-data directory, and loads the models available to your account.

The server listens on http://127.0.0.1:4141 by default.

3. Confirm it is ready

In another terminal:

curl -fsS http://127.0.0.1:4141/readyz
curl -fsS http://127.0.0.1:4141/v1/models

4. Send a request

Choose a model ID returned by /v1/models, replace MODEL_ID, and call the OpenAI-compatible endpoint:

curl http://127.0.0.1:4141/v1/chat/completions \
  -H "content-type: application/json" \
  -d '{
    "model": "MODEL_ID",
    "messages": [
      {"role": "user", "content": "Explain why Rust is useful for API gateways."}
    ]
  }'

That is the complete local path: authenticate once, start one process, and point your client at it.

Connect your client

The most common setup difference is whether the client expects /v1 in its base URL: