cowagent

CowAgent (formerly chatgpt-on-wechat) is an open-source super AI assistant and agent harness. According to its README, it proactively plans tasks, controls the computer and external services, creates and runs Skills, builds a personal knowledge base and long-term memory, forms multi-agent teams, and evolves through use. It is positioned as a reference implementation of Agent Harness engineering. Architecture and capabilities - Planning: decomposes complex tasks and executes them step by step, looping over tools until the goal is reached. - Multi-Agent: builds teams of agents, each with its own role, model, skills, and knowledge, collaborating in a shared conversation. - Memory: a three-tier architecture (context, daily, core) with automatic Deep Dream distillation and hybrid keyword plus vector retrieval. - Knowledge: auto-curates structured knowledge into a Markdown wiki and builds an evolving knowledge graph with visual browsing. - Evolution: reviews conversations to improve skills, follow up on unfinished tasks, and consolidate memory and knowledge. - Skills: one-click install from the Skill Hub, GitHub, or ClawHub, or create custom skills through natural-language conversation. - Tools: built-in file I/O, terminal, browser, scheduler, memory retrieval, web search and more, with native MCP integration. - Multimodal: support for text, images, voice, and files, including recognition, generation, and delivery. Models and channels CowAgent supports many mainstream LLM providers, including Claude, GPT, Gemini, DeepSeek, Qwen, GLM, Kimi, MiniMax, Doubao, ERNIE, MiMo, LinkAI, and custom endpoints. Chat, vision, image generation, ASR/TTS, and embeddings can each be routed to a different vendor, configured from the Web console. A single agent instance can serve multiple channels in parallel, including Web console, Telegram, Slack, Discord, WeChat, Feishu/Lark, DingTalk, WeCom, QQ, and WeChat Official Account. Deployment and usage A one-line installer handles dependencies, configuration, and startup on Linux/macOS, Windows PowerShell, or Docker. After startup, the Web console at localhost:9899 is the hub for chatting, configuring models, connecting channels, and installing skills. A cow CLI manages service control, status, logs, updates, skill installation, and browser automation. A desktop client for macOS and Windows is also offered. Community and licensing The project is MIT licensed and states it is intended for technical research and learning. It warns that agent mode consumes substantially more tokens than regular chat and has access to the local operating system, so it should only be deployed in trusted environments. The README notes the project was previously named chatgpt-on-wechat and that the old GitHub URL redirects automatically.