
Open-source personal AI assistant...
Keeping your own data and running work on a schedule is where QwenPaw differs from hosted chat products. Local deployment leaves the workspace on your machine, and the console on a local port configures models, channels, and skills. Multi-agent mode gives each role its own memory, then collaboration skills split the work. ReMe memory turns conversations into readable, editable, searchable Markdown instead of leaving them only in a chat window. Scheduled jobs can push news digests or reports. It fits people and small teams who will install it themselves and use DingTalk or Feishu as the front door. Install scripts on locked-down corporate machines may need manual environment setup.
QwenPaw is an open-source personal AI assistant from the AgentScope team, short for Qwen Personal Agent Workstation. It runs locally or in the cloud. One instance can host multiple agents, each with its own workspace, memory, skills, and history, and they can message each other. Channels include DingTalk, Feishu, WeChat, Discord, Telegram, iMessage, and QQ, plus a console, terminal UI, and desktop app. ReMe memory stores conversations and files as editable Markdown. Skills cover schedules, documents, the browser, and news. The software is Apache-licensed and can run locally without a cloud API.
Difficulty: Intermediate
Multi-agent workspaces
Create several agents in one instance, each with its own workspace, config, memory, skills, and history, and let them message each other.
Multi-channel messaging
One instance connects DingTalk, Feishu, WeChat, Discord, Telegram, iMessage, and QQ, and you can also chat in the console, terminal, and desktop app.
Editable long-term memory
ReMe turns chats and files into readable, editable, searchable, linked Markdown instead of keeping them only in the current chat.
Skills and scheduled jobs
Built-in skills cover schedules, PDFs and office files, news digests, and file reading. A plugin market and MCP add external tools.
Local model runtime
A built-in local runtime works offline without an API key, and you can switch to Ollama, LM Studio, or a cloud provider.
Scheduled news digests
A person schedules a news digest and the same instance posts it to DingTalk or Feishu.
Editing code in a local project
A developer reads, edits, diffs, and checks project files in the workspace while data stays on their computer.
Working through office files
Hand a PDF, Word file, spreadsheet, or deck to the matching skill to read, rewrite, or convert, then download the result.
It is a personal assistant installed in your own environment. Chat can happen in DingTalk, Feishu, or WeChat, or in the local console. Multiple agents keep separate memory and skills, and scheduled jobs follow your timetable. The repository says local deployment keeps data on your machine. If you attach a cloud model, you still enter the provider key yourself.
After install, run initialization, start the app, and open the local console in a browser to configure a model. The console chats, toggles channels, manages sessions and schedules, edits the persona, imports skills, and manages MCP and multiple agents. DingTalk or Feishu still needs channel setup from the docs. If a corporate network blocks the install script, install manually.
The repository credits the AgentScope team and builds on AgentScope, AgentScope Runtime, and ReMe. The license is Apache, so you can read and modify the source. Install paths include pip, a script, source, and a hosted platform. Free here means the software can run locally; cloud model providers bill separately.
Real reviews and feedback from users