bbd2api

A lightweight Flask reverse proxy that bridges the Claude Messages API and OpenAI Chat Completions API to backboard.io's assistant/thread API — with tool calling, thread pooling, and sub-5s response times.

Features

• Dual API compatibility — /v1/messages (Claude) and /v1/chat/completions (OpenAI)
• Tool calling — prompt-injection approach; compact format (~800 chars vs ~68KB for full schemas)
• Multi-turn tool continuity — tool_result messages route back to the correct thread automatically
• Thread pool — pre-warms N threads at startup; zero cold-start on new conversations
• Global assistant — single assistant reused across all requests; eliminates per-request creation overhead
• System prompt stability — strips dynamic tags (<system-reminder> etc.) before hashing to prevent cache misses
• Model routing — auto-detects provider from model name (claude→anthropic, gpt→openai, gemini→google, etc.)

Quick Start

git clone https://github.com/N1nEmAn/bbd2api.git
cd bbd2api
pip install -r requirements.txt
cp .env.example .env
# edit .env and set BBD_API_KEY
python server.py

Configuration

Variable	Default	Description
BBD_API_KEY	(required)	API key(s) — comma-separated for multi-key rotation: key1,key2,key3
BBD_UPSTREAM	https://app.backboard.io/api	Upstream API base URL
HOST	0.0.0.0	Listen address
PORT	10088	Listen port
THREAD_TTL	1800	Thread cache TTL in seconds
DEBUG	1	Enable debug logging (0 to disable)

API Endpoints

Method	Path	Description
POST	/v1/messages	Claude Messages API (streaming + sync)
POST	/v1/chat/completions	OpenAI Chat Completions API
GET	/v1/models	List available models
GET	/health	Health check
GET	/debug/state	Show assistant/thread/pool state
POST	/debug/clear	Clear thread cache

Usage with Claude Code / OpenAI clients

# Claude-style
curl http://localhost:10088/v1/messages \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-4-6","messages":[{"role":"user","content":"Hello"}],"stream":true}'

# OpenAI-style
curl http://localhost:10088/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o","messages":[{"role":"user","content":"Hello"}]}'

Or point any OpenAI-compatible client (LiteLLM, Continue, etc.) at http://localhost:10088.

Architecture

Client (Claude/OpenAI API)
        │
        ▼
   bbd2api proxy (Flask, port 10088)
        │  • thread pool (pre-warmed)
        │  • tool prompt injection
        │  • tool_result continuity (tool_use_id → thread mapping)
        │  • SSE format translation
        ▼
  backboard.io API
  /assistants/{aid}/threads/{tid}/messages

Tool Calling Flow

1. Client sends request with tools list
2. Proxy injects compact tool format instructions into the user message
3. Model responds with JSON: {"tool":"name","args":{...}}
4. Proxy translates to tool_use content blocks, stores tool_use_id → thread_id
5. Client executes tool, sends back tool_result
6. Proxy detects tool_result, looks up thread via tool_use_id, sends result to same thread
7. Model continues reasoning and produces final answer

License

CC BY-NC-SA 4.0 — Non-commercial use only. If you use or adapt this project, you must credit the original author (N1nEmAn) and distribute your changes under the same license.

Commercial use is prohibited without explicit written permission from the author.

Disclaimer

See DISCLAIMER.md.

---

Community

The author endorses and supports the LINUX DO community.

---

bbd2api（中文文档）

将 Claude Messages API 和 OpenAI Chat Completions API 转接到 backboard.io 助手/线程 API 的轻量级 Flask 反向代理，支持工具调用、线程池预热，响应时间低于 5 秒。

功能特性

• 双 API 兼容 — /v1/messages（Claude）和 /v1/chat/completions（OpenAI）
• 工具调用 — prompt 注入方案，极简格式（~800 字符 vs 完整 schema 的 ~68KB）
• 多轮工具连续性 — tool_result 消息自动路由回正确的 thread，对话不断链
• 线程池 — 启动时预热 N 个 thread，新对话零冷启动
• 全局 assistant — 单一 assistant 复用，消除每次请求创建 assistant 的开销
• 系统提示稳定性 — hash 前剥离动态标签（<system-reminder> 等），防止缓存失效
• 模型路由 — 根据模型名自动识别 provider（claude→anthropic，gpt→openai，gemini→google 等）

快速开始

git clone https://github.com/N1nEmAn/bbd2api.git
cd bbd2api
pip install -r requirements.txt
cp .env.example .env
# 编辑 .env，填入 BBD_API_KEY
python server.py

配置说明

变量	默认值	说明
BBD_API_KEY	（必填）	API Key，逗号分隔多个 Key 可启用轮换：key1,key2,key3
BBD_UPSTREAM	https://app.backboard.io/api	上游 API 基础地址
HOST	0.0.0.0	监听地址
PORT	10088	监听端口
THREAD_TTL	1800	Thread 缓存过期时间（秒）
DEBUG	1	开启调试日志（0 关闭）

接口列表

方法	路径	说明
POST	/v1/messages	Claude Messages API（流式 + 同步）
POST	/v1/chat/completions	OpenAI Chat Completions API
GET	/v1/models	获取可用模型列表
GET	/health	健康检查
GET	/debug/state	查看 assistant/thread/pool 状态
POST	/debug/clear	清除 thread 缓存

调用示例

# Claude 风格
curl http://localhost:10088/v1/messages \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-4-6","messages":[{"role":"user","content":"你好"}],"stream":true}'

# OpenAI 风格
curl http://localhost:10088/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o","messages":[{"role":"user","content":"你好"}]}'

也可以将任意兼容 OpenAI 的客户端（LiteLLM、Continue 等）指向 http://localhost:10088。

架构说明

客户端（Claude/OpenAI API）
        │
        ▼
   bbd2api 代理（Flask，端口 10088）
        │  • 线程池（预热）
        │  • 工具 prompt 注入
        │  • tool_result 续接（tool_use_id → thread 映射）
        │  • SSE 格式转换
        ▼
  backboard.io API
  /assistants/{aid}/threads/{tid}/messages

工具调用流程

1. 客户端发送含 tools 的请求
2. 代理将极简工具格式说明注入用户消息
3. 模型输出 JSON：{"tool":"name","args":{...}}
4. 代理转换为 tool_use content block，并记录 tool_use_id → thread_id
5. 客户端执行工具，返回 tool_result
6. 代理检测到 tool_result，通过 tool_use_id 找回原 thread，将结果发送进去
7. 模型基于结果继续推理，输出最终回答

多 Key 轮换

BBD_API_KEY=key1,key2,key3

• Round-robin 轮换：每次请求取下一个 key
• 自动跳过：某个 key 连续失败 3 次暂时跳过
• 自动恢复：成功一次即重置失败计数

许可证

CC BY-NC-SA 4.0 — 仅限非商业用途。使用或改编本项目时，必须注明原作者（N1nEmAn），并以相同许可证分发衍生作品。

未经作者书面许可，禁止商业用途。

免责声明

见 DISCLAIMER.md。

---

社区

作者认可并支持 LINUX DO 社区。