Personal AI Agent You Can Host, Extend, and Actually Control – CoPaw

CoPaw is a free, open-source, OpenClaw-style personal AI assistant you run on your own machine or in the cloud. It connects to messaging platforms you already use (like Discord, iMessage, etc) and operates as a single agent across all of them. You chat in your preferred app, and CoPaw will do the rest.

The project comes from the AgentScope team and is licensed under Apache 2.0. It sits somewhere between a self-hosted chatbot and a lightweight autonomous agent platform. You can use it for daily content digests, file management, creative drafts, and research tasks, or you can write custom Skills that turn it into whatever you need.

Features

• Built-in cron scheduler: Runs scheduled tasks and reminders to any connected channel.
• Custom Skills: Loads user-written Skills from your workspace directory automatically.
• Long-term memory: Stores context and personalization data under your control, locally or in the cloud.
• Local model backends: Runs LLMs on your hardware via llama.cpp (cross-platform), MLX (Apple Silicon), or Ollama (cross-platform, requires Ollama service).
• Docker support: Ships a pre-built image (agentscope/copaw:latest) on Docker Hub with persistent volume support.
• ModelScope one-click cloud: Forks a Studio instance for users who prefer not to install anything locally.
• Alibaba Cloud ECS deployment: Provides a one-click deployment link for running CoPaw on cloud infrastructure.
• Social content digests: Pulls daily highlights from Reddit, Zhihu, Xiaohongshu, and summarizes Bilibili or YouTube content.
• Desktop file management: Reads, summarizes, and organizes local files on request through chat.
• Email and calendar contacts: Extracts contacts and newsletter content from email and calendar sources.

Use Cases

• Social media digests: Aggregates and summarizes hot posts from Reddit, Zhihu, and YouTube daily.
• Automated newsletters: Receive compiled email and calendar updates directly in your preferred chat app.
• Overnight drafting: Provide a goal before bed and receive a completed draft the next morning.
• Desktop organization: Reads local documents, summarizes files, and organizes folders upon request.
• Research tracking: Build a personal knowledge base from automated tech and AI news alerts.

How to Use It

Installation

Option 1: pip (recommended if you manage Python yourself)

pip install copaw
copaw init --defaults
copaw app

Open http://127.0.0.1:8088/ in your browser to access the Console.

Option 2: One-line installer (no Python required)

macOS / Linux:

curl -fsSL https://copaw.agentscope.io/install.sh | bash

Windows (PowerShell):

irm https://copaw.agentscope.io/install.ps1 | iex

After the installer finishes, open a new terminal and run:

copaw init --defaults
copaw app

Option 3: Docker

docker pull agentscope/copaw:latest
docker run -p 8088:8088 -v copaw-data:/app/working agentscope/copaw:latest

Config, memory, and Skills are stored in the copaw-data volume. Pass API keys via -e VAR=value or --env-file .env.

Option 4: Install from source

git clone https://github.com/agentscope-ai/CoPaw.git
cd CoPaw
pip install -e .

For dev dependencies (tests and formatting):

pip install -e ".[dev]"

To build the Console frontend yourself:

cd console && npm ci && npm run build

Then run copaw app from the project root.

Local Model Setup

CoPaw runs local LLMs without any API key. Install the backend extra, download a model, then start the server.

copaw models download Qwen/Qwen3-4B-GGUF
copaw models
copaw app

You can also manage models from Console → Models.

API Key Configuration (Cloud LLMs)

There’re three ways to set your key:

1. During init: Run copaw init (without --defaults) and follow the LLM provider prompts.
2. Console UI: Go to Settings → Models → select a provider → enter the API Key → activate.
3. Environment variable: Set DASHSCOPE_API_KEY in your shell or in a .env file in the working directory.

For tools that need additional keys (e.g., TAVILY_API_KEY for web search), add them in Console → Settings → Environment variables.

CLI Command Reference

Install script options (macOS / Linux):

Install script options (Windows PowerShell):

Upgrading on any platform: re-run the original install command. CoPaw handles the rest.

Adding Channels

After starting copaw app, open Console → Channels to connect DingTalk, Feishu, QQ, Discord, or iMessage. Each channel has its own credential requirements documented at copaw.agentscope.io/docs/channels. CoPaw routes messages from all connected channels to the same agent, so you can switch platforms without reconfiguring the assistant.

Writing a Custom Agent Skill

Drop a Python file into your CoPaw workspace Skills directory. CoPaw loads it automatically on the next restart. The Skills system has no lock-in. Your Skills are plain Python files with no proprietary decorators required. Combine a Skill with the built-in cron scheduler to build a repeating agentic task.

Pros

• Self-hostable: Data, config, and memory stay on your hardware by default.
• One agent, many channels: A single installation handles DingTalk, Feishu, QQ, Discord, and iMessage simultaneously.
• Offline capable: The llama.cpp and MLX backends run entirely on-device, no API key required.
• Three deployment modes: Local machine, Docker, and cloud (ModelScope or Alibaba Cloud ECS).

Cons

• Hardware requirements: Local model execution demands significant RAM and processing power.
• Setup complexity: Beginner users might find custom skill creation difficult.
• Cloud dependencies: You need external API keys if your machine cannot run local models.

Related Resources

• CoPaw Official Documentation: Full reference covering installation, channels, Skills, memory, and CLI.
• AgentScope: The underlying multi-agent framework that CoPaw builds on.
• ReMe: A companion memory system from the AgentScope team, relevant if you want to understand CoPaw’s memory architecture.
• ModelScope CoPaw Studio: One-click cloud setup for running CoPaw without local installation.
• Ollama: The local model service CoPaw supports as an optional backend.
• llama.cpp: The cross-platform local inference engine CoPaw uses for the llamacpp extra.
• 7 Best OpenClaw Alternatives for Safe & Local AI Agents
• 7 Best & Open-source Claude Cowork Alternatives

FAQs

Q: Does CoPaw require a paid API subscription to work?
A: No. CoPaw runs entirely without a cloud API key if you use a local model backend (llama.cpp, MLX, or Ollama).

Q: Can I run CoPaw on a Windows machine?
A: Yes. The one-line PowerShell installer handles setup on Windows, and llama.cpp is the recommended local model backend for cross-platform use. MLX is Apple Silicon-only and will not work on Windows.

Q: What happens to my data and configuration if I uninstall?
A: Running copaw uninstall removes the application but leaves your configuration and data in the working directory. Running copaw uninstall --purge removes everything, including stored memory and Skills.

Q: How do I connect CoPaw to Discord?
A: Start CoPaw with copaw app, open the Console at http://127.0.0.1:8088/, go to Channels, and select Discord. You will need a Discord bot token from the Discord Developer Portal. The full setup steps are in the Channels documentation.

Q: Can multiple people use the same CoPaw instance?

A: CoPaw is designed as a personal assistant, not a multi-tenant service. If you deploy it on cloud infrastructure and expose it to others, anyone with access to the Console URL can control the agent.

Q: Is CoPaw the same as GitHub Copilot or any Microsoft product?
A: No. CoPaw is a completely separate open-source project from the AgentScope team (Alibaba Group research). It has no connection to GitHub Copilot, Microsoft, or OpenAI.

Q: How do Skills work technically?
A: Skills are Python files you place in your CoPaw workspace directory. CoPaw scans and loads them automatically on startup. Each Skill can define functions the agent calls in response to user requests or scheduled cron triggers. No registration, manifest file, or proprietary decorator is required.