Index / JavaClaw · updated Oct 5, 2026
JavaClaw
ClawRunr/JavaClaw · healthy · rank 33 of 50 by stars
A Java/Spring Boot reimplementation of OpenClaw built as a showcase for JobRunr background jobs and Spring AI. It targets the JVM enterprise crowd with a modular plugin architecture, guided onboarding, and local-first privacy.
Facts
Repository
Runtime
Posture
Community
Security breakdown
Composite 55 / 100 · how these are scored
higher is safer
higher is safer
higher is safer
higher is safer
higher is riskier
Model access
Read from the repository, not written by a model · 63 files examined
2 providers · runs locally
Pinned models
gpt-4o released 2024-05-13 — 2 years old (from the public model catalogue)
Pin last edited 20 days ago
Decision
Why choose JavaClaw over OpenClaw?
Why choose this
- Native JVM/Spring stack fits enterprise Java shops
- JobRunr-powered durable background jobs with built-in dashboard
- Enforced module boundaries via Spring Modulith and ArchUnit tests
Tradeoffs
- Far smaller community and channel/skill ecosystem
- Heavy JVM footprint versus OpenClaw's leaner runtime
- No official releases yet; early-stage project started as a demo
Best fit
- Java/Spring developers wanting an OpenClaw-style agent on the JVM
- Teams already using JobRunr or Spring AI
- Self-hosters who want guided onboarding and a web dashboard
Avoid if
- You need a lightweight, fast-booting agent (JVM startup is heavy)
- You want a mature plugin ecosystem and large community
- You need strong sandboxing around shell/file access
Evidence is solid from the README and detailed recent commits showing active development, but community sentiment data is thin — the Reddit matches are unrelated noise, so sentiment and adoption signals are uncertain.
AI layer reviewed Oct 5, 2026 · how this is written
Star activity
755 stars today
Overview
JavaClaw is a Java 25 / Spring Boot 4 reimplementation of the OpenClaw personal AI assistant, originally created as a demo for JobRunr and now positioned as a community-driven Java agent platform. It acts as a local control plane: a Spring Modulith core (base/ and app/) handles the agent loop, task management, and configuration, while channel plugins (Telegram, Discord, WhatsApp via wacli, WebSocket chat) and tool plugins (Brave search, Playwright browser automation) self-register via Spring Boot auto-configuration. LLM access goes through Spring AI with pluggable providers — OpenAI, Anthropic (including Claude Code OAuth), Google GenAI, and local Ollama — selectable during a guided onboarding flow.
Its standout differentiator is durable background job processing via JobRunr (with a dashboard on :8081), letting the agent schedule one-off, delayed, and cron-recurring tasks stored as human-readable Markdown files. Recent commits show active hardening: ArchUnit-enforced module boundaries, token-by-token WebSocket streaming, multi-model provider support with an agents management page, and careful security fixes like refusing WhatsApp group chats to prevent group members from instructing a shell-capable agent.
Compared to OpenClaw, JavaClaw trades ecosystem breadth and runtime lightness for JVM familiarity and enterprise-grade structure. It keeps data local (embedded H2, file-backed workspace) with no telemetry, but the agent has unrestricted shell and file access with no sandboxing layer, so it carries the same high shell-risk profile as other local agents. There are no official releases yet, so expect a moving target.