ClawRunr/JavaClaw

What it solves

JavaClaw provides a way to run a personal AI assistant locally on your own hardware, ensuring data privacy while giving the agent the ability to interact with your machine and various communication platforms. It solves the problem of fragmented AI tools by acting as a central control plane that can manage tasks, execute system commands, and communicate across different channels.

How it works

Built with Java 25 and Spring Boot, JavaClaw connects to an LLM provider of your choice (OpenAI, Anthropic, or local via Ollama). It uses a "workspace" directory to store its system prompt, environment context, and memory as Markdown files. The assistant extends its capabilities through a plugin-based architecture for communication channels (like Discord, Telegram, and WhatsApp) and a skill system where new abilities can be added simply by dropping Markdown files into a folder or adding JAR packages to the classpath. It also integrates the Model Context Protocol (MCP) to connect to external tool servers and uses JobRunr to handle background and scheduled tasks.

Who it’s for

It is designed for Java developers and power users who want a self-hosted AI agent that can automate local system tasks, manage a personal to-do list, and be accessible via their favorite chat apps.

Highlights

• Multi-Channel Access: Interact with your agent via a built-in web UI, Telegram, Discord, or WhatsApp.
• Local Control: Ability to read/write files and execute bash commands directly on the host machine.
• Extensible Skills: Add new capabilities at runtime using SKILL.md files or pre-packaged JARs.
• Flexible LLM Support: Switch between cloud providers and local models (Ollama) during a guided onboarding process.
• Dynamic Tool Discovery: Uses a search-based pattern to surface only the most relevant tools to the LLM, preventing prompt bloat.