zhayujie/cowagent

CowAgent (originally known as chatgpt-on-wechat) is an AI-powered assistant framework written in Python that lets you deploy a capable, autonomous AI agent across multiple Chinese and international messaging platforms. The core problem it solves is that most large language model (LLM) chat experiences are purely reactive, you ask a question and get an answer. CowAgent goes further by giving the AI the ability to plan multi-step tasks, access files and run terminal commands, browse the web, and remember past conversations persistently. When you set it up, the agent connects to your chosen messaging platform, WeChat (personal or official account), Feishu (Lark), DingTalk, enterprise WeChat, QQ, or even a simple web interface, and listens for messages. When you send a request like "summarize this PDF and email a draft to my colleague," the agent breaks that into steps, calls the relevant built-in tools (file reader, browser, scheduler), and completes the work without you micromanaging each step. All of this happens 24 hours a day on your personal computer or a server. The system supports a wide range of LLM backends, including DeepSeek, OpenAI, Claude, Gemini, MiniMax, Qwen, and GLM. You can switch models in the configuration file. Memory is layered into core memory, daily notes, and longer-term knowledge graphs so the agent can recall relevant context from weeks ago when you revisit a topic. A Skills system lets you extend the agent's capabilities: install community-contributed skills from the Skill Hub, import from GitHub with one command, or instruct the agent to create a new skill through conversation. The tech stack is pure Python (3.7-3.13), deployable on Linux, macOS, or Windows. Docker is supported for setups where you want to skip the manual Python environment configuration. A CLI tool called cow manages starting, stopping, and updating the service.