{"repo":"ATaC-team/ATaC","free":true,"listed":false,"github":"https://github.com/ATaC-team/ATaC","clone":"git clone https://github.com/ATaC-team/ATaC.git","description":"Record and Replay AI Agent Trajectories.","language":"Python","stars":299,"topics":["agent-infra","agents","mcp","skills"],"license":"MIT","category":"ai-agents","readme_excerpt":"English 中文 中文 ATaC是一个面向 Agent 工作流的工具运行时，帮助 Agent 通过编排图代码复用资产。 它主要提供三类能力： - 统一工具注册：把本地 Agent 工具和 MCP 工具纳入同一个运行时 - Agent 自编排 graph：让 Agent 生成的 graph 代码可以直接复用这些已注册能力 - 一键运行：加载并执行这些 graph 工作流 把工具注册到 ATaC，让 Agent 用代码把这些工具组织成流程，并逐步沉淀为可复用、可维护的 graph 工作流。 🤖 最小接入示例 1. 注册 Agent 内置工具和 MCP 工具 推荐做法： - 本地工具直接使用原生 langchain core.tools.tool - MCP 工具直接使用官方 langchain mcp adapters.client.MultiServerMCPClient - 统一通过 service.register langgraph tools(...) 注册到 AtacService 2. 在 graph 代码里调用 ATaC SDK 🚀 快速开始 1. 安装 2. Python SDK 3. 示例 - bootstrap.py - tool call graph.py - agent workflow.py - run graph demo.sh - run graph sdk.py --- English ATaC is a tool runtime for agent workflows. It provides three core capabilities: - unified tool registration - agent-authored graph workflows that reuse registered tools - one-command graph execution Register tools with ATaC, let agents organize them into workflows as code, and gradually turn repeated tasks into reusable, maintainable graph workflows. 🤖 Minimal Integration 1. Register Built-In and MCP Tools Recommended path: - use the native langchain core.tools.tool decorator for local tools - use the official langchain mcp adapters.client.MultiServerMCPClient for MCP tools - register everything through service.register langgraph tools(...) 2. Call the ATaC SDK from Graph Code 🚀 Quick Start 1. Install 2. Run via Python SDK 3. Examples - bootstrap.py - tool call graph.py - agent workflow.py - run graph demo.sh - run graph sdk.py","default_branch":null,"files":null,"tree":[],"storefront":"/r/ATaC-team","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/ATaC-team/ATaC/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}