Alibaba open-sources CoPaw, an easy-to-run personal assistant across chat apps

Alibaba just open-sourced CoPaw, a personal AI assistant built to bridge local intelligence with your everyday chat applications. It provides a visual web interface out of the box, runs seamlessly on your own hardware, and talks directly through the messaging platforms you already use.

Unlike traditional command-line automation setups, it focuses on zero-code operations and pain-free configuration while keeping all your sensitive personal data on your local disk.

What CoPaw can do

At its core, CoPaw acts as an automated multi-channel dispatcher. You can schedule recurring workflows without writing code, such as generating daily summaries from Reddit, social feeds, or video platforms, and pushing neat digests directly to your chat groups.

Its connectivity covers DingTalk, Lark, QQ, WeChat, Discord, Telegram, and iMessage through a single instance. You talk to your assistant in whatever chat app you prefer, and it executes tasks behind the scenes.

Beyond basic chat, CoPaw incorporates a unified file workspace to parse, edit, and convert documents like PDF, Word, and Excel. It also reads code repositories, conducts diff reviews, and runs project tests.

Extensibility is another major highlight. CoPaw supports one-click skill imports from ClawHub and features hot-swappable MCP modularity across prompts, hooks, and tools. Its three-layer ReMe memory architecture combined with local vector search keeps track of personal context and preferences over long horizons.

How to get it running

CoPaw is designed for true offline privacy. It natively integrates with local inference runtimes including Ollama, llama.cpp, and MLX on Apple Silicon, alongside support for LM Studio and over a dozen cloud LLM APIs. The team even provided tailored QwenPaw-Flash models spanning 2B, 4B, 9B, 27B, and 35B-A3B parameters.

Security guardrails are wired straight into the kernel level, featuring sandbox isolation, tool guards, file guards, and permission scanners to block destructive commands before execution.

Getting started takes just one shell command on macOS, Linux, or Windows, which auto-configures virtual environments and dependencies. If you prefer managing Python manually, running pip install qwenpaw followed by the default initialization fires up the web console at port 8088.

Why it matters

If you want an autonomous personal bot that bridges local open-source models with your actual messaging workflow, CoPaw is currently one of the easiest ways in. Packing a native Web UI, built-in MCP tooling, and domestic enterprise chat connectors into a single install script makes personal agent orchestration genuinely accessible to non-programmers.

It puts direct pressure on fragmented agent frameworks by offering a complete end-to-end desktop experience. That said, orchestrating multi-agent tasks on consumer-grade local hardware still demands considerable memory headroom, so testing lighter quantized models first is your safest bet.