CowAgent Rebuilds a 46,500-Star WeChat Bot Into a Self-Evolving Open-Source Agent
CowAgent (formerly chatgpt-on-wechat) is a self-evolving open-source AI agent with long-term memory, a skills marketplace, and support for 10+ LLMs across 12 messaging platforms
PRO- Rebranded:
chatgpt-on-wechatis now CowAgent, a full open-source Agent Harness with 46.5k GitHub stars. - Self-evolving memory: Three-tier memory (context, daily, core) with a nightly "Deep Dream" distillation pass that consolidates long-term knowledge automatically.
- Skills marketplace: One-click install from Skill Hub; custom skills can be created through natural-language conversation with no code.
- Broad model and channel support: 10+ LLM providers (OpenAI, Claude, Gemini, DeepSeek, Qwen, etc.) across 12 channels including WeChat, Telegram, Slack, and Discord.
- Latest release v2.1.6: Adds sub-agents for parallel task delegation, reasoning-effort controls, and a pluggable memory vector backend.
- Free to self-host: MIT licensed, one-line install on Linux/macOS/Windows/Docker; managed hosting available via LinkAI.
CowAgent is the project formerly known as chatgpt-on-wechat, now officially rebranded and rebuilt into something far more ambitious: a full Agent Harness. With over 46,500 GitHub stars and nearly 10,000 forks, it has become one of the most popular open-source personal AI agent frameworks in the world. The core pitch is simple but powerful: a single agent that runs 24/7 on your laptop or server, connects to any major LLM, plugs into every messaging platform you use, and actually gets smarter over time.
From WeChat Bot to Agent Harness
This project was previously named chatgpt-on-wechat and is now officially CowAgent. The rename reflects a genuine architectural shift. Version 2.0.0 was a major upgrade to a super Agent assistant with multi-step task planning, long-term memory, and the Skills framework. What started as a WeChat chatbot wrapper is now a complete agentic runtime.
The platform has evolved beyond a simple chatbot into a more autonomous agent capable of planning complex tasks, maintaining long-term memory, and invoking external tools to complete workflows. The concept of an "Agent Harness" is central here: it is the layer that sits between your LLM and the real world, handling planning, tool dispatch, memory, and channel routing so the model can focus on reasoning.
How It Actually Works
CowAgent is a complete Agent Harness: messages flow in through Channels; the Agent Core plans and reasons over memory, knowledge, and the available tools and skills; Models generate the response, which is sent back through the originating channel. Every layer is decoupled and independently extensible.
The key capabilities that make CowAgent more than a chatbot wrapper:
- Task Planning: Decomposes complex tasks and executes them step by step, looping over tools and skills until the goal is reached.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
