zhayujie/CowAgent

Open-source super AI assistant & Agent Harness. Plans tasks, runs tools and skills, self-evolves with memory and knowledge. Multi-agent, multi-model, multi-channel. Lightweight, extensible, one-line install. (formerly chatgpt-on-wechat)

What it solves

CowAgent is an open-source AI assistant designed to move beyond simple chat by proactively planning tasks, controlling computers and external services, and evolving its own capabilities over time. It provides a unified framework (an "Agent Harness") to deploy a personal AI agent that can manage long-term memory, structured knowledge, and a wide array of tools across various messaging platforms.

How it works

CowAgent uses a decoupled architecture where messages enter through Channels (like Telegram or Slack) and are processed by the Agent Core. The core reasons over available tools and memory to plan a sequence of actions. It supports a wide range of LLM providers for different modalities (chat, vision, image generation, ASR/TTS).

Key technical components include:

  • Three-tier Memory: A system that moves information from short-term context to daily memory and finally into a long-term MEMORY.md file via a nightly "Deep Dream" distillation process.
  • Knowledge Base: An auto-curated Markdown wiki and knowledge graph that organizes information by topic rather than time.
  • Skills & Tools: Atomic tools (like bash, browser automation, and file I/O) are composed into higher-level "Skills" which can be installed from a hub or created via natural language.
  • MCP Integration: Native support for the Model Context Protocol to integrate external tool servers.

Who it’s for

It is built for users who want a persistent, 24/7 AI assistant running on their own hardware or server, as well as developers who want to build and deploy complex agentic workflows across multiple IM platforms.

Highlights

  • Multi-Agent Teams: Ability to create teams of agents with distinct roles and knowledge that collaborate in a shared conversation.
  • Self-Evolution: Automatically reviews conversations to improve skills and follow up on unfinished tasks.
  • Broad Integration: Supports a vast array of channels including WeChat, Telegram, Slack, Discord, and Feishu.
  • Extensible Tooling: Built-in browser automation, terminal access, and a one-click Skill Hub for expanding capabilities.
  • Model Agnostic: Easy swapping of providers (Claude, GPT, Gemini, DeepSeek, etc.) via a web console.

Related

  • Project
    fy-agent/fyagent

    FyAgent is a cross‑platform desktop app that consolidates the configuration of AI‑assistant tools (model/provider selection, skill/plugin installation, prompt and memory editing, and service credentials) into a single local UI, helping users manage multiple AI agents consistently.

  • Project
    Alpha-Dojo/DojoAgents

    DojoAgents is an open‑source AI‑agent framework for personal investing. It runs a large‑language‑model‑driven “Agent Loop” that fetches market data, parses news, executes Python‑based calculations, and stores reusable analysis skills. A FastAPI backend and React dashboard let users view portfolio metrics, get daily market overviews, run news‑impact analyses, and even diagnose a portfolio from a screenshot. Installable via `uv pip install dojoagents`, it works with any LLM API (OpenAI, Gemini, Anthropic, local Ollama, etc.) and can push automated insights to chat apps.

  • Project
    deepagent-ltd/deepagent-code

    An AI coding workspace for long-running tasks that uses a document graph and persistent project memory to maintain coherence across complex software engineering workflows.

  • Project
    the-open-agent/openagent

    An open-source personal AI assistant platform that combines LLMs, RAG, and autonomous agent loops into a single, self-hostable binary.

  • Project
    itwanger/PaiAgent

    PaiAgent is an open‑source, enterprise‑grade platform that lets users visually compose AI workflows with drag‑and‑drop nodes. It supports multiple large‑model providers via Spring AI, offers two execution engines (a classic DAG engine and a LangGraph4j state‑graph engine), includes a declarative “Skills” system for LLM‑driven function calls, and provides real‑time SSE debugging. The stack is React + TypeScript on the front‑end and Spring Boot 3 + Java 21 on the back‑end, with MySQL (and optional MinIO) for persistence.