QwenPaw
A self-hosted AI assistant with persistent memory and chat app access, using local models through llama.cpp, Ollama or LM Studio.

QwenPaw is a personal AI assistant for people who want an agent on their own computer or server. It retains conversation history and a personal knowledge base across sessions.
- Local models: The built-in llama.cpp runtime runs QwenPaw-Flash models without API keys or cloud services; Ollama and LM Studio are also compatible.
- Persistent memory: Working context, verbatim history and a ReMe knowledge base turn conversations and resources into linked Markdown files you can read, edit and search.
- Shared access: The web console, terminal interface and chat channels share agents, memory and sessions, with connections including DingTalk, Lark and Discord.
- Tasks and extensions: Schedule recurring work and add document, browser or news capabilities through skills, plugins and MCP tools; independent agents can work in parallel with their own memory and skills.
- Execution controls: Operating-system sandboxes, tool-call checks, sensitive-file protection and skill scanning restrict agent actions, with access policies that allow, deny or request human approval.
QwenPaw is open source under Apache 2.0 and runs on macOS, Linux and Windows. Local inference runs entirely on your machine. Optional cloud providers, including OpenAI, Anthropic and Google Gemini, require API keys and process model requests remotely. AgentScope Platform also offers a cloud deployment.