JClaw - Java-based Enterprise AI Assistant

JAVA FIRST. NO BLOAT. PURE POWER.

Open source under the MIT License. Use it, change it and share it, for any purpose. See License for the details.

What is JClaw?

JClaw lets you run your own AI assistants. We call each assistant an agent.

You talk to your agents in your web browser, or in Slack, Telegram or WhatsApp. An agent can search the web, read your documents and do work on a schedule. It also remembers what matters to you.

JClaw runs on your own computer or server. You choose the AI model behind each agent. It can be a cloud service, such as OpenAI or OpenRouter, or a model that runs on your own computer.

See it in action

https://github.com/user-attachments/assets/b355102e-5eff-49d5-8e4e-5e4fff47ddb2

A 2-minute demo of three JClaw features: tasks, the prompt library, and deep research. If you hear no sound, click the speaker icon to turn on the voiceover. You can also watch it on YouTube.

What you can do

• 💬 Chat with your agents. Ask questions, give them work and read their answers.
• 📱 Reach them where you already talk. Connect an agent to Slack, Telegram or WhatsApp.
• ⏰ Put work on a schedule. A task runs at the time you choose, such as a daily news briefing. JClaw sends the result to the place you pick.
• 🔔 Set reminders. JClaw sends you a message at the time you ask for.
• 📝 Keep a prompt library. Save the instructions you use often, and use them again at any time.
• 🔬 Research in depth. An agent can search many sources, check them and write you a full report.
• 🌐 Read the web. Agents open web pages for you. They can also read a whole website in the background.
• 📄 Read your files. Agents read your documents, PDFs and images.
• 🎙️ Talk out loud. Speak to an agent and hear its answer in voice mode.
• 🧠 Remember what matters. Agents keep useful facts about you and your work. You can see and change what they remember.
• 👥 Share out big jobs. An agent can start helper agents that work at the same time.
• 🧩 Teach it new skills. Add skills, and connect other apps and services, so your agents can do more.
• 🗂️ Get small apps built for you. An agent can turn a request you repeat into a small web app.
• 🛡️ Stay in control. Agents ask for your approval before risky actions, such as running commands on your computer.

Fast, even with smaller models

In our own tests, we gave JClaw, Claude Cowork, Hermes and OpenClaw the same tasks and prompts. JClaw was often 5 to 10 times faster. Its answers were also better, even on less powerful models.

Most of this comes from context engineering. This means that JClaw is careful about what it puts in front of the AI model at each step:

• Only what the task needs. JClaw lists each skill and each large connected service by its name and a short summary. The agent loads the full details only when it uses them.
• Shorter tool output. A long tool result reaches the model whole once. After that, JClaw can send a shorter version. The agent can fetch the full text again if it needs it.
• Long chats stay in budget. When a conversation nears the model's limit, JClaw summarizes the older parts. It keeps the recent turns word for word.

A smaller prompt is faster to read, costs less and gives the model less to get wrong.

To measure JClaw's agents yourself, use the test sets in evals/. They check that an agent picks the right tools, returns valid structured answers and answers only from its sources. DEVELOPMENT.md shows you how to run them.

Easy to set up and run

We also compared how you set up and run JClaw, Hermes and OpenClaw. JClaw was the easiest of the three:

• One command installs it. The installer downloads JClaw and starts it. If you need Java, it offers to download it for you.
• The rest happens in your browser. You choose your password, connect an AI model, set up agents and connect channels in the JClaw web app. There is no setup wizard to run in a terminal.
• Chat apps in a few clicks. On the Channels page, fill in the bot's details, choose the agent, then save and switch it on. There is no gateway to restart: a Telegram or Slack bot answers at once. Each bot can run a different agent, and a Test button checks its token. For WhatsApp, you can also link a phone by scanning a QR code in your browser. If a chat app needs a public web address, one switch can make one for you with Tailscale.
• No files to edit. For everyday use, you change settings on pages in the app, not in configuration files.
• Your own password protects it. You sign in to the web app with the password you chose.
• Safe updates. Update from the app or with one command. JClaw backs up your database first. If the new version does not start, JClaw goes back to the old one.

Get started

1. Install JClaw

On macOS or Linux, open a terminal and run this command:

curl -fsSL https://github.com/tsukhani/jclaw/releases/latest/download/install.sh | sh

On Windows, open PowerShell and run this command:

irm https://github.com/tsukhani/jclaw/releases/latest/download/install.ps1 | iex

The installer downloads JClaw and starts it. JClaw needs Java 25. If your computer does not have it, the installer offers to download it for you. On Windows, JClaw also needs Git Bash or WSL (the Windows Subsystem for Linux).

2. Open JClaw

Open http://localhost:9000 in your web browser. The first time, JClaw asks you to choose a password for your account.

3. Connect an AI model

Go to Settings → Providers → LLM Providers and set up one provider:

• For a cloud service, paste in the API key that the service gives you. An API key is a secret code that lets JClaw use your account.
• For a model on your own computer, install Ollama or LM Studio first. INSTALL.md shows you how.

If the provider shows 0 models, click the round-arrow button on its models row (Discover models from provider). Then add the models you want.

Then tell your first agent to use that provider:

1. Open Agents and click the main agent.
2. Choose your provider in Default Provider.
3. Choose a model in Default Model.
4. Save the agent.

4. Say hello

Open Chats and send a message to the main agent.

The first time you sign in, JClaw offers you a short tour of the app. The User Guide, under Help in the sidebar, explains every page.

Everyday commands

Run these in a terminal:

To do this	Run this
See if JClaw is running	jclaw status
Stop JClaw	jclaw stop
Restart JClaw	jclaw restart
Update to the newest version (your data stays)	jclaw upgrade
Remove JClaw and everything it stores	jclaw uninstall

You can also update JClaw in the app, from Settings → System → Maintenance.

Learn more

• User guide: how to use each page of JClaw. It is inside the app. Once JClaw runs, open Help → User Guide in the sidebar, or go to http://localhost:9000/guide.
• INSTALL.md: Docker, updates, HTTPS, optional extras and every install option.
• DEVELOPMENT.md: for the team that builds JClaw.
• Release notes: what changed in each version.

Get help

• Found a bug or have a question? Open a GitHub issue, or email [email protected].
• Found a security problem? Email [email protected]. Do not open a public issue for it.
• Want to contribute code? Open a pull request on GitHub. CONTRIBUTING.md explains how.

Where JClaw comes from

JClaw takes ideas from three earlier projects: OpenClaw, JavaClaw and Hermes. All of JClaw's code is new. It shares no code with them. JClaw is written in Java, and it is built to be light and fast.

License

JClaw is open source under the MIT License. You may use, change and share it, for any purpose, commercial use included.

• The third-party components in JClaw's releases keep their own licenses. THIRD-PARTY-NOTICES.md lists each one.
• The JClaw name and logo are trademarks of Abundent Sdn Bhd. TRADEMARK.md explains how you may use them.
• Contributions are welcome under the same license; see CONTRIBUTING.md.

---

Built with ☕ Java and ❤️ by the Abundent crew.