What is this?
A Go gateway that sits in front of FreeBuff and exposes OpenAI-compatible and Anthropic-compatible chat endpoints, translating requests into the FreeBuff wire protocol, with a built-in Svelte admin dashboard.
Is this a proxy server I run myself, or a hosted API?
It is a self-hosted proxy server (binary or Docker container) that you run on your own machine; it exposes /v1/chat/completions, /v1/models, and an /admin dashboard.
How do I get it running?
Configure the .env file from .env.example, then start the Go backend (or use the included Dockerfile/compose setup) to bring up the gateway and the embedded dashboard at /admin.
Do I need FreeBuff auth tokens to use this?
Yes — it relies on FreeBuff credentials. Setting AUTHTOKENS puts it in pooled or hybrid mode; leaving AUTHTOKENS empty switches it to bridge mode where each request must carry its own token.
Is this project still maintained?
No — the repository is explicitly marked deprecated and unmaintained, so updates, bug fixes, or compatibility with FreeBuff changes are not guaranteed going forward.
Who would use this?
Developers who want to talk to FreeBuff using standard OpenAI or Anthropic API clients instead of FreeBuff's native wire protocol.
What are the three operating modes?
Pooled mode uses a shared token pool (AUTHTOKENS set, bridge disabled); bridge mode relays each request's own token (AUTHTOKENS empty); hybrid mode (the default when AUTHTOKENS is set) uses the pool for APIKEYS credentials and relays any other credential as a bridge token.
How does the freebucks metering work?
Freebucks are charged once per session-hour at session start, refunded if the session is deleted early, and refilled on a Pacific-midnight schedule, following FreeBuff's wire prices map.
Are there built-in rate limits?
Yes, the default .env.example ships with a safe-mode anti-ban preset capping usage at 30 requests per minute and 1500 requests per day, Pacific time.
What does the admin dashboard show?
It's a Svelte single-page app embedded directly in the Go binary, served at /admin, for managing and monitoring the gateway.
What are similar projects?
opencode2api, ccx, bifrost, grok2api, vm2api, casbin-gateway, workbuddy2api-panel, claude-relay-service
Which topics is it in?
llm-proxy, api-gateway, reverse-proxy, openai-compatible, deprecated, proxy, docker-compose, anthropic, dashboard, rest-api, openai, go