GhostBrain — Gemini to API

GhostBrain resurrects a free Gemini web account as real infrastructure. The web app gives you a powerful model behind a chat page; GhostBrain lifts it out and serves it as a proper API — with the polish that implies: secure session binding, multi-account pools, SSE streaming and persistence.

Design decisions

The fragile part of any web-account bridge is the session. GhostBrain treats it as a first-class citizen: cookies are stored encrypted, sessions are health-checked and rotated before they die instead of after, and accounts are pooled so no single identity absorbs all the traffic. When a session does die, the pool quietly marks it and keeps serving — your client never sees a 401.

Concern	Approach
Session security	Encrypted cookie vault, no plaintext secrets on disk.
Reliability	Background health probes + automatic rotation.
Compatibility	OpenAI /v1/chat/completions and Anthropic /v1/messages.
Streaming	Native SSE relay, token-for-token.
Language	Python — async I/O end to end.

Where it fits my stack

GhostBrain is the Python sibling of GLM Free API, and both are upstreams my OmniRouter happily load-balances. Deployed together on free tiers — Hugging Face Spaces, Railway, Render — they give me a multi-model API fleet whose monthly bill is exactly zero. That combination is what my Hermes Stack ships as a one-click package.

Responsible use

Free-tier bridges live on someone else's generosity. GhostBrain rate-limits itself and rotates accounts precisely so it stays polite — use it like a guest, not a looter.