GitHub radar
CowAgent: Self-Evolving AI Assistant with Skills and Memory
An open-source AI agent that plans tasks, controls your computer and services, installs skills from a hub, builds a personal knowledge base, and improves itself over time — running 24/7 on your own machine.
CowAgent (formerly chatgpt-on-wechat) is an open-source AI agent that runs on your machine or server around the clock. It plans and decomposes complex tasks step by step, uses built-in tools (browser, file I/O, terminal, scheduler, web search, MCP integration), installs Skills from a public hub or from GitHub, and builds a structured personal knowledge base and knowledge graph. A three-tier memory architecture automatically distills long-term knowledge through 'Deep Dream' conversations. Self-Evolution reviews past sessions to improve skills and consolidate memory. Supports any major LLM — Claude, GPT, Gemini, DeepSeek, Qwen, and more — switchable from the web console. Connects to Web, Telegram, WeChat, Feishu, DingTalk, Slack. One-line installer for Linux/macOS/Windows; Docker and cloud options available.
Why a vibe-coder should care
Unlike a session-based assistant that forgets everything after you close the tab, CowAgent maintains long-term memory, learns from past conversations, and stays running between tasks. If you want an agent that knows your context and improves over time — rather than starting fresh every chat — this is the architecture for it.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Deploy CowAgent from https://github.com/zhayujie/CowAgent — follow the README, use the one-line installer, configure it to use Claude (ask me for the key if needed), start the web console and show me it's working.
Runs on a regular laptop — Python plus an API key for at least one supported model (Claude, GPT, Gemini, etc.).
Open on GitHub▌ More finds