Open-source personal AI assistant CowAgent released, supports multi-agent collaboration and self-evolution

CowAgent is an open-source super AI assistant and agent framework (Agent Harness) that 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 grows through self-evolution in use. Its lightweight design needs just one install script for dependencies, configuration and startup, supporting local, Docker and server deployments, with the web console on port 9899 by default. The memory system uses a three-layer architecture (conversation context, daily memory, core memory), distilled nightly via Deep Dream with hybrid keyword and vector retrieval; the knowledge base is auto-organized into a Markdown wiki with a visualized knowledge graph. It bundles over ten built-in tools (file read/write, terminal, browser, scheduled tasks, web search) and natively integrates the MCP protocol; skills install in one click from Skill Hub, GitHub and more, or are created via natural-language chat. Models include Claude, GPT, Gemini, DeepSeek, Qwen, GLM, Kimi and other major vendors, with chat, vision, image generation, speech recognition/synthesis and embeddings routable to different vendors and switchable in one click in the web console. Channels cover Web, WeChat, Feishu, DingTalk, WeCom, QQ, Telegram, Slack and more, with one agent instance serving multiple channels in parallel. For developers connecting assistants to Chinese office and messaging tools, and teams wanting a local 24/7 personal assistant, this framework offers a ready-to-use complete solution.