pingmike2/freebuff2api-wokers

Freebuff/Codebuff free models as an OpenAI-compatible API — Docker-first, zero deps, bilingual web admin panel (/admin), per-account SOCKS5 egress.

What it solves

This project removes the hassle of using freebuff/codebuff's free AI models. Normally you'd have to manage login tokens, keep track of daily usage limits (called Freebucks), and switch accounts when one runs out. This service wraps those free models into a standard OpenAI-compatible API, so you can plug any OpenAI or Anthropic client into it and let it handle all the messy account, quota, and session management for you.

How it works

The service runs as a lightweight server (a single file with no dependencies) that accepts OpenAI-style requests. Behind the scenes, it keeps a pool of freebuff accounts. When you make a chat request, it picks an account that still has credits, uses a session (which lasts about an hour) to call the free model, and returns the result in the standard format. It also includes a web admin panel (at /admin) where you can add or remove account tokens, set a separate proxy for each account, see how many Freebucks each account has left, and adjust API settings. All changes apply instantly without restarting. It supports both OpenAI and Anthropic API formats, and has a health check endpoint for uptime monitoring.

Who it's for

This is for developers, tinkerers, or small teams who want to use free AI models (like DeepSeek, Kimi, GLM, etc.) through their favorite OpenAI-compatible tools, without having to build and maintain their own account rotation, quota tracking, and API gateway. It's especially helpful if you need to manage multiple free accounts to get more daily usage, or if you want a simple way to expose those models to existing clients.

Highlights

  • Single-file, dependency-free code that can run on Docker or a VPS.
  • Web admin panel to manage accounts, proxies, API keys, and view Freebucks balances in real time.
  • Multi-account auto-switching: when an account hits its daily limit, it automatically cools down and switches to another.
  • Smart session reuse: sessions last about an hour and are pinned to the same account to maximize credit usage.
  • Compatible with both OpenAI (/v1/chat/completions) and Anthropic (/v1/messages) APIs.
  • Health check endpoint for uptime monitoring.
  • Docker deployment is recommended; Cloudflare Worker deployment is discouraged because it risks account bans.

Related

  • Project
    LeiWang1999/ZYNQ-NVDLA

    ZYNQ‑NVDLA is an open‑source project that ports NVIDIA’s Deep Learning Accelerator (NVDLA) to Xilinx Zynq FPGA boards. It provides RTL, Vivado IP, Linux drivers, pre‑built ARM64 libraries, and a Tengine‑Lite backend so you can run quantised neural‑network models (e.g., ResNet‑18, YOLOX‑nano) on the FPGA‑based accelerator with low latency.

  • Project
    tile-ai/tilelang-puzzles

    TileLang Puzzles is a collection of 10 hands-on exercises that teach you TileLang, a domain-specific language for writing high-performance deep learning kernels. Progressing from trivial examples to modern kernels like GEMM and FlashAttention, it gives you a practical way to learn kernel design principles with reference solutions for every challenge.

  • Project
    pleaseai/shunt

    A transparent LLM gateway that allows you to route Claude Code requests to alternative AI providers, enabling you to use different models while keeping your agent's tools and workflow intact.

  • Project
    Leavesfly/TinyAI

    A full-stack AI framework built entirely in Java that provides a complete pipeline from low-level tensor computation to LLMs, AI agents, and embodied intelligence.