Open-source AI Coding Agent framework with SDK, Skills, and MCP integration
Project description
Cody
开源 AI Coding Agent 框架 — 构建、定制和部署你自己的 AI 编程 Agent。
Cody 提供构建 AI 编程 Agent 所需的完整基础设施:28 个工具、11 个内置技能、MCP/LSP 集成、子 Agent 编排、会话管理和安全体系。你可以用 SDK 将它嵌入任何 Python 应用,也可以直接用 CLI/TUI/Web 开箱即用。
为什么选择 Cody?
| 痛点 | Cody 怎么解决 |
|---|---|
| 想自建 AI 编码工具,但从零造轮子太重 | 28 个工具 + 安全体系 + Sessions 全现成,专注你的业务逻辑 |
| Claude Code / Cursor 不够灵活,想定制 Agent 行为 | Skills 系统 + 权限控制 + 多模型切换,完全可控 |
| 绑定单一模型厂商,切换成本高 | 多模型支持(Claude、GPT、Gemini、DeepSeek、智谱 GLM、通义千问) |
| 商业产品无法审计、无法私有部署 | 开源 MIT,代码在你手里,可审计、可定制、可离线部署 |
快速开始
方式一:SDK 嵌入(推荐)
pip install cody-ai # 仅 4 个核心依赖
from cody import AsyncCodyClient
async with AsyncCodyClient(workdir="/path/to/project") as client:
# 让 AI 执行编码任务
result = await client.run("创建一个 FastAPI hello world 应用")
print(result.output)
# 多轮对话(自动创建 session)
r1 = await client.run("创建 Flask 应用")
await client.run("添加 /health 端点", session_id=r1.session_id)
# 流式输出
async for chunk in client.run_stream("解释这段代码"):
print(chunk.content, end="")
SDK 直接调用核心引擎(in-process),无需启动任何服务。详细文档:SDK 使用指南
方式二:CLI 开箱即用
pip install cody-ai[cli]
# 配置模型
cody config setup
# 执行任务
cody run "创建一个 FastAPI hello world 应用"
# 交互对话
cody chat
# 全屏终端
cody tui
方式三:Web 界面
pip install cody-ai[web]
cody-web --dev # 开发模式(含 Vite HMR)
框架能力一览
28 个内置工具
| 分类 | 工具 |
|---|---|
| 文件 I/O | read_file, write_file, edit_file, list_directory |
| 搜索 | grep, glob, search_files, patch |
| Shell | exec_command |
| 子代理 | spawn_agent, get_agent_status, kill_agent |
| MCP | mcp_call, mcp_list_tools |
| Web | webfetch, websearch |
| LSP | lsp_diagnostics, lsp_definition, lsp_references, lsp_hover |
| 文件历史 | undo_file, redo_file, list_file_changes |
| 任务管理 | todo_write, todo_read |
| 用户交互 | question |
| 技能 | list_skills, read_skill |
Agent Skills 开放标准
兼容 Agent Skills 开放标准(Claude Code、Cursor、GitHub Copilot 等 26+ 平台采用)。你的 Skills 可以跨平台复用。
---
name: git
description: Git 版本控制操作。处理 git 仓库时使用。
metadata:
author: cody
version: "1.0"
---
# Git 操作
AI 代理的使用说明...
11 个内置技能: git, github, docker, npm, python, rust, go, java, web, cicd, testing
三层优先级: .cody/skills/(项目)> ~/.cody/skills/(用户)> 内置
多模型支持
| 提供商 | 模型示例 |
|---|---|
| Claude | claude-sonnet-4-0, claude-opus-4-0 |
| OpenAI | openai:gpt-4, openai:gpt-4-turbo |
google:gemini-pro |
|
| DeepSeek | deepseek:deepseek-coder |
| 智谱 GLM | glm-4(需配置 model_base_url) |
| 阿里通义 | qwen-coder-plus(需配置 model_base_url) |
| 阿里百炼 | qwen3.5(需配置 coding_plan_key) |
| 任何 OpenAI 兼容 API | 通过 model_base_url 配置 |
集成能力
- MCP 集成 — 通过 stdio JSON-RPC 连接外部 MCP 服务器(GitHub、数据库等)
- LSP 代码智能 — Python (pyright)、TypeScript (tsserver)、Go (gopls)
- 子代理系统 — 孵化专业代理(code/research/test),asyncio 并发执行
- 上下文管理 — 接近 token 限制时自动压缩对话,智能文件分块
安全体系
- 工具级权限控制(allow/deny/confirm)
- 路径遍历保护 + 危险命令检测
- 审计日志(SQLite 持久化)
- 速率限制(滑动窗口)
- 文件修改 undo/redo
四种使用方式
Cody 的核心是 AI 编程引擎(cody/core/),以下四种方式共享同一个引擎:
| 方式 | 适用场景 | 安装 |
|---|---|---|
| SDK | 嵌入到你的应用/平台/工具链 | pip install cody-ai |
| CLI | 终端中快速执行任务 | pip install cody-ai[cli] |
| TUI | 全屏终端交互(Textual) | pip install cody-ai[tui] |
| Web | 浏览器界面 + HTTP API | pip install cody-ai[web] |
# 一次性安装全部
pip install cody-ai[all]
配置
# 交互式配置向导(推荐,首次使用时自动触发)
cody config setup
# 或手动设置环境变量
export CODY_MODEL_API_KEY='sk-ant-...'
# 使用 OpenAI 兼容 API(如智谱 GLM)
export CODY_MODEL='glm-4'
export CODY_MODEL_BASE_URL='https://open.bigmodel.cn/api/paas/v4/'
export CODY_MODEL_API_KEY='sk-...'
# 阿里云百炼 Coding Plan
export CODY_MODEL='qwen3.5'
export CODY_CODING_PLAN_KEY='sk-sp-xxxxx'
详细配置:配置文件详解
开发
# 从源码安装
git clone https://github.com/CodyCodeAgent/cody.git
cd cody
pip install -e ".[dev]"
# 运行核心 + SDK 测试(588 个)
uv run pytest tests/ -v
# Web 后端测试(85 个)
PYTHONPATH=. uv run pytest web/tests/ -v
# Web 前端测试(33 个)
cd web && npx vitest run
# Lint(必须零告警)
uv run ruff check cody/ tests/ web/
文档
入门
框架开发
参考
许可证
MIT License
致谢
基于以下优秀项目构建:
最后更新: 2026-03-07 | 版本: 1.7.4
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file cody_ai-1.10.3.tar.gz.
File metadata
- Download URL: cody_ai-1.10.3.tar.gz
- Upload date:
- Size: 195.8 kB
- Tags: Source
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/6.1.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
97976ed854896dbe6a68cf99fde0211ee776390d959ce4763faae4201baf4fa3
|
|
| MD5 |
786d958af0dcfb3b79c96669ac65d109
|
|
| BLAKE2b-256 |
5be55a7a0cccd3c169208a75cfc7632a1bb8a7e7aaaa9def70cb4d6b33c883f7
|
Provenance
The following attestation bundles were made for cody_ai-1.10.3.tar.gz:
Publisher:
python-publish.yml on CodyCodeAgent/cody
-
Statement:
-
Statement type:
https://in-toto.io/Statement/v1 -
Predicate type:
https://docs.pypi.org/attestations/publish/v1 -
Subject name:
cody_ai-1.10.3.tar.gz -
Subject digest:
97976ed854896dbe6a68cf99fde0211ee776390d959ce4763faae4201baf4fa3 - Sigstore transparency entry: 1114887603
- Sigstore integration time:
-
Permalink:
CodyCodeAgent/cody@eec414d1de51ca8241721e3b7fa3342da5532c8d -
Branch / Tag:
refs/tags/v1.10.3 - Owner: https://github.com/CodyCodeAgent
-
Access:
public
-
Token Issuer:
https://token.actions.githubusercontent.com -
Runner Environment:
github-hosted -
Publication workflow:
python-publish.yml@eec414d1de51ca8241721e3b7fa3342da5532c8d -
Trigger Event:
push
-
Statement type:
File details
Details for the file cody_ai-1.10.3-py3-none-any.whl.
File metadata
- Download URL: cody_ai-1.10.3-py3-none-any.whl
- Upload date:
- Size: 170.8 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/6.1.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
be25386b2af92d3fc73e11212814415aa016d0d063e0e7c22bec4159fbc8d9b8
|
|
| MD5 |
aa5710dc241e9afcc81d6cd23a2d21dd
|
|
| BLAKE2b-256 |
8b6c2e5fd41e0a0c4aadf46a7fb2796140a315ac8eaa4d79d77b5ec492ba7cb9
|
Provenance
The following attestation bundles were made for cody_ai-1.10.3-py3-none-any.whl:
Publisher:
python-publish.yml on CodyCodeAgent/cody
-
Statement:
-
Statement type:
https://in-toto.io/Statement/v1 -
Predicate type:
https://docs.pypi.org/attestations/publish/v1 -
Subject name:
cody_ai-1.10.3-py3-none-any.whl -
Subject digest:
be25386b2af92d3fc73e11212814415aa016d0d063e0e7c22bec4159fbc8d9b8 - Sigstore transparency entry: 1114887610
- Sigstore integration time:
-
Permalink:
CodyCodeAgent/cody@eec414d1de51ca8241721e3b7fa3342da5532c8d -
Branch / Tag:
refs/tags/v1.10.3 - Owner: https://github.com/CodyCodeAgent
-
Access:
public
-
Token Issuer:
https://token.actions.githubusercontent.com -
Runner Environment:
github-hosted -
Publication workflow:
python-publish.yml@eec414d1de51ca8241721e3b7fa3342da5532c8d -
Trigger Event:
push
-
Statement type: