Skip to content

dgqyushen/command-code-adapter

BranchesTags

Repository files navigation

Command Code Adapter

中文 | English

中文

将 Command Code API 暴露为兼容 OpenAI Chat Completions、Anthropic Messages 和 OpenAI Responses 格式的适配器。

支持流式（SSE）和非流式响应，附带 Web 管理面板。

快速开始

# 安装依赖
poetry install

# 配置 API Key
export CC_ADAPTER_CC_API_KEY=user_your_key_here

# 启动服务
poetry run python -m cc_adapter

服务启动后访问 http://localhost:8080，管理面板在 http://localhost:8080/admin。

Docker

# 构建并运行
docker build -t cc-adapter .
docker run -p 8080:8080 -e CC_ADAPTER_CC_API_KEY=user_your_key_here cc-adapter

# 或使用 docker-compose（推荐）
# 编辑 .env 文件配置 CC_ADAPTER_CC_API_KEY，然后：
docker compose up -d

配置

环境变量 默认值 说明
CC_ADAPTER_CC_API_KEY — Command Code API Key（必填）
CC_ADAPTER_CC_BASE_URL https://api.commandcode.ai CC API 地址
CC_ADAPTER_HOST 0.0.0.0 监听地址
CC_ADAPTER_PORT 8080 监听端口
CC_ADAPTER_LOG_LEVEL INFO 日志级别
CC_ADAPTER_LOG_FORMAT console 日志格式：console 或 json
CC_ADAPTER_ADMIN_PASSWORD — 管理面板密码（留空则无需认证）
CC_ADAPTER_ACCESS_KEY — API 访问密钥（留空则无需认证）
CC_ADAPTER_DEFAULT_MODEL deepseek/deepseek-v4-flash 管理面板 Playground 默认模型
CC_ADAPTER_HTTP_MAX_CONNECTIONS 200 HTTP 连接池最大连接数
CC_ADAPTER_HTTP_MAX_KEEPALIVE_CONNECTIONS 50 HTTP 连接池最大 Keepalive 连接数
CC_ADAPTER_HTTP2 false 启用 HTTP/2
CC_ADAPTER_ZDR true 发送 x-cmd-zdr: 1 请求头（零数据留存）
CC_ADAPTER_WEB_SEARCH_PROVIDER — 设为 deepseek 时，将 Anthropic web_search 请求转发到 DeepSeek
CC_ADAPTER_DEEPSEEK_API_KEY — DeepSeek API Key，用于 web_search 转发
CC_ADAPTER_DEEPSEEK_ANTHROPIC_URL https://api.deepseek.com/anthropic DeepSeek Anthropic 兼容端点
CC_ADAPTER_WEB_SEARCH_MODEL — web_search 转发使用的 DeepSeek 模型；留空则使用请求中的模型

也可通过 .env 文件配置（参考 .env.example）。

日志

日志格式通过 CC_ADAPTER_LOG_FORMAT 控制（默认 console）。

Console 格式（默认）— 人眼可读的结构化单行日志：

07:42:18 INFO  app.start        base=https://api.commandcode.ai port=8080
07:42:31 INFO  openai.request   model=deepseek-v4-flash stream=true message_count=3 tools=true req=8f3a91c2
07:42:33 WARNING  upstream.retry   reason=empty_response attempt=1 max_attempts=2 req=8f3a91c2
07:42:34 INFO  upstream.usage   model=deepseek-v4-flash input=120 output=834 total=954 elapsed=2.1s req=8f3a91c2
07:42:34 INFO  http.done        method=POST path=/v1/chat/completions status_code=200 elapsed=2.48s req=8f3a91c2

JSON 格式 — 机器解析用结构化格式：

{"event": "http.done", "level": "info", "logger": "cc_adapter.main", "method": "POST", "path": "/v1/chat/completions", "status_code": 200, "elapsed": "2.480s", "request_id": "8f3a91c2", "timestamp": "2026-05-13T07:42:34Z"}

事件列表： app.start、http.done、openai.request、anthropic.request、upstream.error、upstream.retry、upstream.usage、tool.call、auth.failed、admin.login.failed、admin.config.updated、admin.verify_key

脱敏： 敏感字段（authorization、API key、messages、tool 编辑参数等）自动替换为 ***。

使用

OpenAI Chat Completions：

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Hello"}],
    "stream": true
  }'

Anthropic Messages：

curl http://localhost:8080/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: sk-your-key" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

兼容任意 OpenAI SDK：

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8080/v1",
    api_key="not-needed",
)

Reasoning Effort

适配器支持 reasoning_effort 参数，用于控制模型的思考推理强度。

值 说明
"off" 关闭推理输出（响应端过滤 reasoning-delta）
"low" 最小推理
"medium" 中等推理
"high" 较高推理
"xhigh" 高推理
"max" 最大推理
null / 不传 不设推理

支持的取值因模型而异。适配器内置了 MODEL_REASONING_EFFORTS_MAP（来源：CC v0.26.7 客户端数据），当传入的强度值超出模型支持范围时，自动向上最近值（nearest-higher）映射。模型不在映射表中时不设该参数。

模型 支持的值
deepseek/deepseek-v4-* high, max
claude-sonnet-4-6, claude-opus-4-6/7 low, medium, high, xhigh, max
gpt-5.5, gpt-5.4, gpt-5.3-codex low, medium, high, xhigh
gpt-5.4-mini, claude-haiku-4-5 low, medium, high
Qwen/Qwen3.6-Max-Preview, Qwen/Qwen3.6-Plus low, medium, high
stepfun/Step-3.5-Flash low, medium, high

实现：对 reasoning_effort 按模型支持范围进行 clamp（nearest-higher）后透传给 CC API，不注入任何 system prompt。响应端保留 reasoning-delta 的过滤逻辑（"off" 模式下剥离 reasoning_content）。

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Solve 2x+5=13"}],
    "reasoning_effort": "high",
    "stream": true
  }'

运行测试

poetry run pytest

目录结构

cc_adapter/
├── main.py                # FastAPI 应用入口与路由
├── core/                  # 基础设施
│   ├── config.py          #   配置管理（pydantic-settings）
│   ├── errors.py          #   错误处理与状态码映射
│   ├── logging.py         #   日志配置与中间件
│   ├── auth.py            #   认证逻辑（HMAC token）
│   ├── runtime.py         #   运行时单例（config/client/translator）
│   └── utils.py           #   工具函数
├── command_code/          # CC API 客户端
│   ├── client.py          #   HTTP 客户端
│   ├── headers.py         #   请求头构造
│   └── body.py            #   请求体构造
├── providers/             # 协议翻译
│   ├── openai/            #   OpenAI 兼容
│   │   ├── router.py      #   POST /v1/chat/completions
│   │   ├── models.py      #   数据模型
│   │   ├── request.py     #   OpenAI → CC
│   │   └── response.py    #   CC → OpenAI
│   ├── anthropic/         #   Anthropic 兼容
│   │   ├── router.py      #   POST /v1/messages
│   │   ├── models.py      #   数据模型
│   │   ├── request.py     #   Anthropic → CC
│   │   └── response.py    #   CC → Anthropic
│   └── shared/            #   共享工具
│       ├── model_mapping.py  # 模型名映射
│       └── tool_mapping.py   # 工具参数名映射
├── catalog/               # 数据目录
│   └── models_data.py     #   模型列表
└── admin/                 # Web 管理面板
    └── ...                #   管理界面

English

Quick Start

# Install dependencies
poetry install

# Configure API Key
export CC_ADAPTER_CC_API_KEY=user_your_key_here

# Start server
poetry run python -m cc_adapter

Once started, visit http://localhost:8080. The admin panel is at http://localhost:8080/admin.

Docker

# Build and run
docker build -t cc-adapter .
docker run -p 8080:8080 -e CC_ADAPTER_CC_API_KEY=user_your_key_here cc-adapter

# Or use docker-compose (recommended)
# Edit the .env file to set CC_ADAPTER_CC_API_KEY, then:
docker compose up -d

Configuration

Variable Default Description
CC_ADAPTER_CC_API_KEY — Command Code API Key (required)
CC_ADAPTER_CC_BASE_URL https://api.commandcode.ai CC API base URL
CC_ADAPTER_HOST 0.0.0.0 Listen address
CC_ADAPTER_PORT 8080 Listen port
CC_ADAPTER_LOG_LEVEL INFO Log level
CC_ADAPTER_LOG_FORMAT console Log format: console or json
CC_ADAPTER_ADMIN_PASSWORD — Admin panel password (leave blank for no auth)
CC_ADAPTER_ACCESS_KEY — API access key (leave blank for no auth)
CC_ADAPTER_DEFAULT_MODEL deepseek/deepseek-v4-flash Admin Playground default model
CC_ADAPTER_HTTP_MAX_CONNECTIONS 200 HTTP connection pool max connections
CC_ADAPTER_HTTP_MAX_KEEPALIVE_CONNECTIONS 50 HTTP connection pool max keepalive connections
CC_ADAPTER_HTTP2 false Enable HTTP/2
CC_ADAPTER_ZDR true Send x-cmd-zdr: 1 header (zero data retention)
CC_ADAPTER_WEB_SEARCH_PROVIDER — Set to deepseek to forward Anthropic web_search requests to DeepSeek
CC_ADAPTER_DEEPSEEK_API_KEY — DeepSeek API key used for web_search forwarding
CC_ADAPTER_DEEPSEEK_ANTHROPIC_URL https://api.deepseek.com/anthropic DeepSeek Anthropic-compatible endpoint
CC_ADAPTER_WEB_SEARCH_MODEL — DeepSeek model used for web_search forwarding; empty means use the request model

You can also configure via a .env file (see .env.example).

Logging

The log format is controlled by CC_ADAPTER_LOG_FORMAT (default: console).

Console format (default) — human-readable single-line logs:

07:42:18 INFO  app.start        base=https://api.commandcode.ai port=8080
07:42:31 INFO  openai.request   model=deepseek-v4-flash stream=true message_count=3 tools=true req=8f3a91c2
07:42:33 WARNING  upstream.retry   reason=empty_response attempt=1 max_attempts=2 req=8f3a91c2
07:42:34 INFO  upstream.usage   model=deepseek-v4-flash input=120 output=834 total=954 elapsed=2.1s req=8f3a91c2
07:42:34 INFO  http.done        method=POST path=/v1/chat/completions status_code=200 elapsed=2.48s req=8f3a91c2

JSON format — structured for machine parsing:

{"event": "http.done", "level": "info", "logger": "cc_adapter.main", "method": "POST", "path": "/v1/chat/completions", "status_code": 200, "elapsed": "2.480s", "request_id": "8f3a91c2", "timestamp": "2026-05-13T07:42:34Z"}

Event names: app.start, http.done, openai.request, anthropic.request, upstream.error, upstream.retry, upstream.usage, tool.call, auth.failed, admin.login.failed, admin.config.updated, admin.verify_key

Redaction: Sensitive fields (authorization, API keys, messages, tool edit parameters) are automatically replaced with ***.

Usage

OpenAI Chat Completions:

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Hello"}],
    "stream": true
  }'

Anthropic Messages:

curl http://localhost:8080/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: sk-your-key" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Compatible with any OpenAI SDK:

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8080/v1",
    api_key="not-needed",
)

Reasoning Effort

The adapter supports the reasoning_effort parameter to control the model's reasoning/thinking intensity.

Value Description
"off" Suppress reasoning output (response-side reasoning-delta filtering)
"low" Minimal reasoning
"medium" Moderate reasoning
"high" High reasoning
"xhigh" Extra high reasoning
"max" Maximum reasoning
null / not set No reasoning effort set

Supported values vary per model. The adapter uses MODEL_REASONING_EFFORTS_MAP (sourced from CC v0.26.7 client data). When a value exceeds a model's supported range, it is clamped to the nearest higher supported value. Models not in the map get no reasoning_effort param.

Model Supported Values
deepseek/deepseek-v4-* high, max
claude-sonnet-4-6, claude-opus-4-6/7 low, medium, high, xhigh, max
gpt-5.5, gpt-5.4, gpt-5.3-codex low, medium, high, xhigh
gpt-5.4-mini, claude-haiku-4-5 low, medium, high
Qwen/Qwen3.6-Max-Preview, Qwen/Qwen3.6-Plus low, medium, high
stepfun/Step-3.5-Flash low, medium, high

Implementation: clamps reasoning_effort to the model's supported range (nearest-higher), then forwards to CC API. No system prompt injection. Response-side filtering strips reasoning-delta events when set to "off".

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Solve 2x+5=13"}],
    "reasoning_effort": "high",
    "stream": true
  }'

Running Tests

poetry run pytest

Directory Structure

cc_adapter/
├── main.py                # FastAPI app entry & routes
├── core/                  # Infrastructure
│   ├── config.py          #   Configuration (pydantic-settings)
│   ├── errors.py          #   Error handling & status code mapping
│   ├── logging.py         #   Logging & middleware
│   ├── auth.py            #   Auth (HMAC token)
│   ├── runtime.py         #   Runtime singletons
│   └── utils.py           #   Utility functions
├── command_code/          # CC API client
│   ├── client.py          #   HTTP client
│   ├── headers.py         #   Request headers
│   └── body.py            #   Request body builder
├── providers/             # Protocol translators
│   ├── openai/            #   OpenAI-compatible
│   │   ├── router.py      #   POST /v1/chat/completions
│   │   ├── models.py      #   Data models
│   │   ├── request.py     #   OpenAI → CC
│   │   └── response.py    #   CC → OpenAI
│   ├── anthropic/         #   Anthropic-compatible
│   │   ├── router.py      #   POST /v1/messages
│   │   ├── models.py      #   Data models
│   │   ├── request.py     #   Anthropic → CC
│   │   └── response.py    #   CC → Anthropic
│   └── shared/            #   Shared utilities
│       ├── model_mapping.py  # Model name mapping
│       └── tool_mapping.py   # Tool param name mapping
├── catalog/               # Data directory
│   └── models_data.py     #   Model listings
└── admin/                 # Web admin panel
    └── ...                #   Admin interface

License

MIT

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

2 stars

Watchers

0 watching

Forks

1 fork
Report repository

Releases

1 tags

Packages

 
 
 

Contributors

Languages