Your free Copilot account, re-routed as a local OpenAI API

What it does

Windows-Copilot-API is an unofficial bridge that logs into your personal Microsoft Copilot account via browser automation, then exposes the chat session as a local REST API. It ships as both a Python library with streaming support and a FastAPI server that mimics the OpenAI /v1/chat/completions format, so existing SDKs work by simply swapping the base URL to localhost. You authenticate once in a visible browser; after that, tokens refresh automatically and your session stays local.

The interesting bit

The project treats Copilot’s consumer web interface as a free, unbilled compute backend, complete with a built-in token-bucket rate limiter to keep your account from getting throttled. Because Copilot’s backend serializes conversations per account, the server deliberately queues parallel requests behind a lock rather than pretending to scale—honest engineering for what is essentially polite scraping.

Key highlights

• Drop-in OpenAI compatibility: point the official openai SDK at http://localhost:8000/v1 and it just works.
• No API keys or credits: uses your existing free Microsoft account; session cookies stay in a local, git-ignored folder.
• Auto session refresh: sign in once via Playwright; the bridge handles token renewal silently after that.
• Self-protecting rate limits: defaults to 12 requests per minute with a burst of 4, returning standard 429 codes to keep upstream happy.
• Cross-platform: runs on Windows, macOS, and Linux under Python 3.9+.

Caveats

• Throughput is intentionally serial: stress tests show 4 concurrent requests is the practical ceiling before Copilot returns 502 errors.
• Only one model is exposed—copilot—and the README benchmarks it at roughly GPT-4 class (40.9% on GPQA Diamond), not a reasoning-tier model.
• This is consumer-account automation, not an official API; it can break if Microsoft changes the web UI or auth flow.

Verdict

Worth a look if you want a free, private LLM backend for personal scripts or local agents without signing up for another API plan. Skip it if you need reliable throughput, SLAs, or anything resembling production infrastructure.