CoPaw: Review & Guide to the Open-Source Personal AI Assistant from Alibaba Tongyi 98

🦞 What is CoPaw

CoPaw, short for Co Personal Agent Workstation, is an open-source personal AI assistant released by Alibaba Cloud's Tongyi Lab. It was officially open-sourced on February 28 (Apache 2.0 license).

It's a lightweight "headless lobster"-style agent framework—in plain terms, it doesn't come with a large model built in; you need to connect an external large model's API to serve as its "brain." It focuses on zero-barrier deployment, smooth localized experience, and proactively working with you, with the goal of being an "all-purpose digital partner" that truly understands you.

CoPaw Features

• Channel Integration: Supports 15+ mainstream platforms including DingTalk, Feishu, QQ, Discord, iMessage, and Telegram. Wherever you send a message, it responds there.

• Dual-Mode Deployment: One-click switch between local (Linux/macOS/Windows) and cloud (ModelScope Studio), with setup taking around 3 minutes.

• Intelligent Memory System: Built on the ReMe memory system, it remembers conversation content and includes a persistent vector store (ReMeFs) that accumulates experience across turns and stores long-term memory.

• Scheduled Task Execution: Supports Cron expressions, enabling automated tasks like generating daily reports, cleaning files, and tracking news.

• File-Based Processing: Configurations and memories are stored as Markdown files (such as SOUL.md, PROFILE.md), making migration and backup easy.

• Skill Extension: Comes with 50+ built-in skills, capable of handling PDF, Excel, Word, and PPT files, as well as managing emails and creating news summaries. Adding new skills doesn't require coding—just configure them.

• Proactive Heartbeat Mechanism: Periodically sends you messages, maintains memories, and executes tasks on its own—it's not just waiting for commands.

• MCP Protocol Support: Through the MCP protocol, it can connect to third-party tools (such as MySQL databases), enabling a closed loop where you can "execute tasks and get results while chatting."

CoPaw Use Cases

Workplace Productivity

• Automatically summarize messages from DingTalk groups into daily reports, then push them to Feishu

• Convert meeting recordings into structured minutes with key points, and extract to-do items on the side

• Monitor industry news sites and push important information to you

Automated Operations

• Scan the downloads folder every day at 9 AM, delete expired files, and notify you via email

• Periodically fetch news, check stock prices, check the weather, and push updates proactively

Content Creation Assistance

• Generate video script outlines based on keywords, and integrate with CapCut API to auto-edit

• Aggregate trending posts from Zhihu and Xiaohongshu, and perform in-depth analysis

Personal Knowledge Management

• Record diet and exercise data, and provide health suggestions

• Control home devices through smart home APIs

Developer Empowerment

• Provides SDKs for quick integration with internal company systems

• Supports private model deployment, such as LLMs dedicated to the financial industry

Pros and Cons of CoPaw

Pros

• Extremely Simple Deployment: One-click Docker setup, running in 3 minutes; also deployable quickly via Alibaba Cloud Compute Nest

• Great Chinese-Language Experience: Specifically adapted for localization, with configuration guides in Chinese and accurate understanding

• Unified Multi-Platform Support: One backend manages all chat apps, no need to switch back and forth

• Privacy Control: With local deployment, data stays on your own machine; you can also connect to local models like Ollama without going online

• Low Extension Barrier: Write a SKILL.md description file to add new skills, no coding required

• High Cost-Performance: Open-source and free, saving significantly compared to tools like Cursor at $20/month

• Proactive Service: Supports scheduled tasks and heartbeat mechanisms, enabling true automation

Cons

• Limited Context Window: When processing large files (over 500 lines), completion quality drops

• Weak Complex Style Support: Complex styles like CSS Grid and animations aren't well supported

• Platform Adaptation Needs Improvement: Channel interaction fluency, such as on Feishu, still has room for improvement, with noticeable latency at times

• Insufficient Multimodal Support: Currently, the conversation window only supports text; images cannot be uploaded

• Opaque Token Consumption: The number of tokens used per operation cannot be viewed in real time

• Weak Concurrent Programming Support: When generating Go code involving goroutines, channels, etc., there may be data race risks

• Skill Ecosystem Still Developing: Compared with OpenClaw, the skill volume and community ecosystem are still under construction

How to Use CoPaw

Quick Deployment (Docker Method, Recommended for Beginners)

# 1. 拉取镜像
docker pull agentscope/copaw:latest
# 2. 运行容器
docker run -p 8088:8088 -v copaw-data:/app/working agentscope/copaw:latest
# 3. 访问控制台
# 浏览器打开 http://127.0.0.1:8088

Key Configuration Steps

Initialization Configuration

copaw init --defaults

Configure the Large Model

On the "Models" page of the console, add an API Key (supports Tongyi Qianwen, DeepSeek, OpenAI, etc.), or download a local model:

copaw models download qwen/qwen3-4b-gguf

Connect Communication Channels

• DingTalk: Create an enterprise internal app, enable the bot, and fill in the Client ID and Secret

• Feishu: Configure long-connection mode (WebSocket), fill in appid and appsecret

• QQ, Discord: Connect via bot mode

Enable Skills and Scheduled Tasks

Enable the skills you need on the "Skills" page, such as PDF processing and news summarization.

Configure scheduled tasks (using Cron expressions) on the "Schedules" page

Common Commands Cheat Sheet

• Initialize: copaw init --defaults

• Start service: copaw app (default port 8088)

• Specify port: copaw app --port 8089

• Download model: copaw models download [model name]

• Upgrade: copaw upgrade or pip install --upgrade copaw

• Check version: copaw --version

CoPaw FAQ

Q1: After deployment, can't access the console (http://127.0.0.1:8088)?

When deploying, remember to open the security group.

Q2: The one-click install script gets stuck?

A: Downloading dependencies can be slow under domestic network conditions. It's recommended to set up a proxy in advance, or just install from source.

Q3: Wrong Python version?

A: CoPaw requires Python 3.10 to 3.13. Use pyenv or conda to switch the version.

Q4: Model invocation fails?

A: Check whether the API Key is filled in correctly, whether the network is reachable, and whether the model name is in the correct case.

Q5: Slow response from the local model?

A: Check whether your machine has enough resources. You can switch to a smaller model, or limit the context length.

Q6: The DingTalk/Feishu bot isn't responding?

A: Check whether the event subscription URL is configured correctly, whether the permissions are enabled, and whether the network can be accessed by the platform's servers. For Feishu, make sure you're not using long connection mode.

Q7: Channel is set up but no messages are received?

A: Check whether the credentials are correct, whether the necessary permissions are enabled, whether the bot has been added to the group chat, and whether the channel shows as "Enabled" in the console.

Q8: Scheduled tasks don't run?

A: Check whether the Cron expression is written correctly, whether the CoPaw service is running, and whether the heartbeat interval and active time window settings conflict.

Q9: How to back up and restore data?

A: Configuration and memory data are stored by default in the ~/.copaw/ directory. To back up, just copy this directory. To restore, overwrite it and restart the service.

Q10: Can CoPaw reply to multiple channels at the same time?

A: Yes! It supports using multiple channels simultaneously, and messages will be automatically routed to the corresponding channel.