Keep your AI assistant where your work already lives.
CoPaw is a deployable personal AI workstation for people who want more than a chat box: real channels, real Skills, real control over memory, models, and ongoing workflows.
Zero-Python setup for local machines
curl -fsSL https://copaw.agentscope.io/install.sh | bashBest for first-time users who want CoPaw running fast.
Browser Console for providers, channels, context, token usage, and chat-state control.
If you want production help or traffic-ready rollout, route deployment through EasyClaw.
A workstation, not just a model wrapper
CoPaw becomes compelling when the runtime, channels, Skills, and operating surface work together. The site should communicate that product shape immediately.
One assistant, wherever you already talk
Connect CoPaw to DingTalk, Feishu, QQ, Discord, iMessage, Telegram, Matrix, Mattermost, and more. The assistant follows your real communication surface instead of forcing a new habit.
Deploy locally, in Docker, or in the cloud
Run CoPaw on your laptop, a private server, or managed infra. Choose the model layer, provider, and deployment shape that fits your security and latency constraints.
Expand behavior without getting locked in
Capabilities come from built-in and custom Skills. Put SKILL.md-based instructions in your workspace, attach scripts when needed, and let CoPaw load them as first-class behavior.
Let it keep working after the chat ends
Scheduled prompts, reminders, digests, and recurring workflows run through heartbeat and cron. CoPaw is designed to stay useful between requests, not only inside a single session.
Private by default, cloud when you choose
Run with local models through llama.cpp, MLX, Ollama, or use hosted providers. Memory, tasks, and channel data remain in your environment unless you decide otherwise.
Operate the whole system from one workstation
The browser Console lets you configure providers, channels, context management, skills, token usage, chat state, and model switching without treating the project like a pile of scripts.
Built with AgentScope. Framed as a product.
CoPaw should feel like its own system. AgentScope matters as a technical backbone, but the buyer-facing story is the workstation: runtime, Console, docs, and deployment paths.
Python core
The core runtime handles agent orchestration, providers, Skills, memory, token tracking, security gating, and channel adapters.
Browser console
A dedicated Console makes configuration, chat control, model switching, and operational visibility usable for humans instead of only maintainers.
Docs as product surface
Docs are not an afterthought. Installation, channel setup, Skills, CLI, and operating patterns are part of the actual user experience.
AgentScope-backed
CoPaw uses AgentScope as a technical foundation, but the product is its own workstation layer with concrete deployment, integration, and day-two workflows.
Meet the assistant in the places that already own your attention
The multi-channel layer is one of CoPaw's clearest product differentiators. That message should be impossible to miss.
Multiple ways to get live, one clear path when you want help
The official product supports script, pip, Docker, and cloud-oriented setups. For deployment-led traffic or managed rollout, point people to your EasyClaw site.
Script install
No Python setup required. The installer handles uv, environment creation, and dependency bootstrapping.
curl -fsSL https://copaw.agentscope.io/install.sh | bashpip install
For teams already comfortable with Python environments and reproducible local development.
pip install copaw
copaw init --defaults
copaw appDocker
A clear path to persistent self-hosted deployment with explicit ports and mounted data volume.
docker pull agentscope/copaw:latest
docker run -p 8088:8088 -v copaw-data:/app/working agentscope/copaw:latestCloud support
ModelScope Studio and cloud deployment flows exist for users who want hosted setup without giving up control of the product surface.
Use ModelScope Studio or a managed deployment partner for production rollout.Need a production-ready rollout or traffic handoff?
Use EasyClaw when you want a guided deployment path instead of stitching together infra and launch support yourself.
Designed for actual repeat use, not one impressive demo
The value of a workstation is cumulative. These are the kinds of loops CoPaw is built to own over weeks and months.
Signal digestion
Track social feeds, AI news, and community chatter, then deliver summaries to the channel you actually check.
Work operating system
Push digests, reminders, and email or calendar follow-ups into your team's daily communication loop.
Creative overnight runs
Describe a goal before you log off and let scheduled workflows return with a draft, brief, or research packet later.
Personal knowledge base
Build a long-lived assistant that keeps context about your tools, preferences, files, and repeat tasks over time.
Desktop bridge
Read files, summarize documents, work with PDFs or Office assets, and coordinate local automation from chat.
Lifestyle automations
Track routines, reminders, and recurring personal check-ins with the same assistant that handles your work surface.
Why CoPaw occupies a different lane
The site needs a clear comparison frame: frameworks are powerful but raw, cloud assistants are polished but closed, and CoPaw sits in the productized middle.
| Capability | CoPaw | Agent shell / framework | Closed cloud assistant |
|---|---|---|---|
| Core focus | Long-lived personal AI workstation | Agent framework or autonomous agent shell | General conversational assistant |
| Channels | Native multi-channel integrations | Usually single surface or API-first | Mostly proprietary app surface |
| Deployment | Laptop, Docker, cloud, self-hosted | Varies by framework and engineering effort | Vendor cloud only |
| Customization | Workspace Skills plus provider and channel config | Code-centric customization | Limited extension hooks |
| Privacy posture | User-controlled environment and model path | Depends on infra choices | Vendor managed by default |
| Day-two operations | Console, docs, token usage, config UI | Often left to builders | Easy, but closed |
Long-form proof that this is a real product surface
You asked to keep the proof-heavy layer. This version keeps it, but rewrites it around CoPaw's product shape instead of making the homepage feel like a legal filing.
Agent core loop
CoPaw is presented as a workstation layer built atop AgentScope rather than a clone of a generic agent shell.
- Uses a productized agent runtime with visible chat, config, tools, and operating workflows.
- Keeps the loop understandable through Console and docs rather than only source code.
- Framework-first systems emphasize generic orchestration primitives.
- Cloud assistants hide the execution loop completely.
Context and memory management
The product stance is explicitly long-lived use, so context management is a feature, not an implementation detail.
- Surfaces memory, context controls, and operating guidance in the product layer.
- Frames retention and compaction as part of day-two ergonomics.
- Typical chat apps reset context frequently or make memory opaque.
- DIY agent stacks leave compaction and persistence mostly to builders.
Skill system
CoPaw treats behavior extension as a user-facing system with workspace-native files, not just plugin APIs for engineers.
- Uses `SKILL.md`-driven behavior with optional scripts and clear trigger descriptions.
- Keeps built-in and custom Skills close to the workspace where the user already operates.
- General frameworks optimize for code extensibility first.
- Closed assistants expose only narrow integrations.
Operational surface
The meaningful differentiator is that CoPaw ships product surface area around the runtime: docs, Console, deployment paths, and channel setup.
- Pairs runtime power with a browser Console, docs, install flows, and concrete channel operations.
- Feels closer to a workstation you can keep than a demo you fork once.
- Frameworks often stop at SDKs and code examples.
- Cloud products give polish but little ownership.
A workstation lens that can keep expanding
The roadmap should feel like compounding leverage, not just a feature wish list.
Large-small model collaboration across privacy-sensitive and high-capability tasks.
Richer multimodal interaction beyond text-only workflows.
Broader channel coverage, stronger ecosystem, and deeper operational tooling.
Latest public release as of March 12, 2026
Native or supported messaging surfaces
Primary deployment modes: local, Docker, cloud
Workstation lens across console, docs, and runtime
Questions people actually ask before they commit
This section should remove ambiguity around what CoPaw is, how it runs, and where deployment support goes.
Build your own assistant. Route deployment help where it belongs.
Explore the docs if you want to self-serve. If you want the traffic or deployment flow to land on your operational site, send people to EasyClaw.