JobRunr Introduces ClawRunr, an Open-Source Java AI Agent

Diogo Carleto · rss · 2026-05-18 Reader 🖥️

Agent Skills Agent Tool Management Agent Context Files (CLAUDE.md / AGENTS.md)Long-Running & Managed Agents MCP (Model Context Protocol)OpenClaw

ClawRunr is JobRunr's open-source Java AI agent that wraps Spring AI's ChatClient with JobRunr's durable, retryable, scheduled job execution, packaging conversational LLM interaction with first-class persistence, recurrence, and monitoring so that long-running agent work behaves like production background jobs rather than ephemeral chat completions.

practical · industry news / product announcement — An InfoQ news article announcing ClawRunr, JobRunr's open-source Java AI agent. The piece is descriptive and informational, walking through stack, architecture, task model, tools, skills, onboarding, provenance, license, and roadmap. It carries an implicit endorsement of treating durable execution (retries, persistence, scheduling, monitoring) as foundational to agent design.

Structure

Part	Role
Announcement and positioning	Introduces ClawRunr as JobRunr's open-source Java AI agent and frames its design pitch: making retries, persistence, scheduling, and monitoring first-class concerns of long-running agent work
Technology stack	Lists Java 25, Spring Boot, Spring AI, Spring Modulith, JobRunr, embedded H2; LLM providers OpenAI, Anthropic, Ollama
Architecture	Describes three Spring Modulith modules (base, app, plugin), Spring AI ChatClient with @Tool registration, Channel interface decoupling message sources, AGENT.md/INFO.md for runtime prompt composition
Task execution model	Tasks stored as Markdown under workspace/tasks/, lifecycle states todo/in_progress/completed/awaiting_human_input, recurring tasks via JobRunr cron with retries and dashboard monitoring
Built-in capabilities	Tools for tasks, shell, files, web scraping; MCP server integration (HTTP + stdio); optional Brave Search and Playwright; opt-in Lucene-based dynamic tool discovery
Skill system	File-based: workspace/skills/<name>/SKILL.md, hot-loaded by the runtime without code or redeploy
Onboarding and configuration	Guided flow selecting provider, credentials, prompt, MCP, channels; persisted to application.yaml; Ollama enables fully local execution
Provenance and license	Began as JavaClaw (Java port of OpenClaw), rewritten by JobRunr team, released under LGPL-3.0
Known issues and roadmap	Playwright first-install restart; web chat weaker than Telegram for recurring task output; roadmap adds Slack/WhatsApp, memory, planning, tools, skills

Key terms

ClawRunr OpenClaw
An open-source Java AI agent built by the JobRunr team that runs on users' own hardware and executes scheduled, recurring, and one-off background tasks while supporting conversational interaction over multiple channels
JobRunr
A Java background job processing library providing scheduling, recurring cron jobs, automatic retries, and a monitoring dashboard; here used as the durable execution backbone for agent tasks
Spring AI ChatClient
Spring AI's abstraction over LLM providers, used as the agent's conversational core; wraps tool registration via @Tool annotations
Channel
An interface in ClawRunr decoupling message sources (built-in web chat, Telegram, Discord plugins) from the agent core
AGENT.md / INFO.md Agent Context Files (CLAUDE.md / AGENTS.md)
Workspace files used to compose the agent's system prompt at runtime: AGENT.md for instructions, INFO.md for environment context
Task lifecycle Long-Running & Managed Agents
States a task moves through (todo, in_progress, completed, awaiting_human_input) with tasks stored as Markdown files under workspace/tasks/
MCP (Model Context Protocol) MCP (Model Context Protocol)
Protocol for connecting external tools to the agent; supported over both streamable HTTP and stdio transports
Dynamic tool discovery Agent Tool Management
Spring AI's Tool Search Tool pattern using Lucene keyword search to surface relevant tools at request time instead of including all tool definitions in every prompt
Skill (file-based) Agent Skills
A capability added by creating a directory under workspace/skills/ with a SKILL.md file inside; loaded by the runtime without code change or redeployment
Spring Modulith module
A modular partitioning approach; ClawRunr uses base (core agent, tasks, tools, channels, config), app (Spring Boot entry, onboarding UI, web chat), and plugin (Brave, Discord, Telegram, Playwright)

Propositions

diagnostic claimAI agent design commonly underserves retries, persistence, scheduling, and monitoring of long-running work.
design claimestablishedA mature background-job library (JobRunr) can serve as the durable execution substrate for an LLM agent.
supports ←The Production Gap: 5 Patterns for Building Long-Running AI Agents · p3 — Generalizes ClawRunr's JobRunr-as-substrate design into a pattern: agents as long-running processes borrowing checkpointing, partial-failure handling, and idempotency from job infrastructure.
extends ←The Production Gap: 5 Patterns for Building Long-Running AI Agents · p3 — Generalizes ClawRunr's specific move (reuse a mature job library for durability) into the principle: model agents as long-running server processes borrowing data-pipeline patterns.
supports ←Article: Implementing Durable Workflows on Postgres Without an External Orchestrator · p1 — Same underlying claim — reuse an already-operated piece of infrastructure as the durable-execution substrate instead of adopting a dedicated engine — applied to Postgres directly instead of a Java background-job library.
architectural claimDecoupling message sources from the agent core via a Channel interface enables multi-surface deployment.
design claimcorroboratedComposing the system prompt from workspace files (AGENT.md, INFO.md) makes agent behavior configurable without code changes.
supports ←Zero to Agent in 30 Minutes: Build a Sports Concierge Agent with Chester Ismay · p3 — Both pieces treat a workspace markdown file (AGENT.md/INFO.md there, CONCIERGE.md here) as the actual configurable control surface for agent behavior — this is a second, independent case of the same pattern.
design claimStoring tasks as Markdown files with explicit lifecycle states makes agent work inspectable and human-editable.
efficiency claimLucene-based dynamic tool discovery reduces prompt bloat by avoiding inclusion of all tool definitions per request.
extensibility claimFile-based skills loaded from a workspace directory enable runtime extensibility without redeployment.
operational claimcontestedRunning against Ollama allows fully local, data-private agent execution.
contradictsGoogle Brings MCP Support to Colab, Enabling Cloud Execution for AI Agents · p2 — Re-asserts local, data-private execution (Ollama) as the goal, against Colab's premise that cloud delegation is the better answer for agent execution.
strategic claimReleasing under LGPL-3.0 with a Java-native stack positions the project for the Java enterprise community.

Arguments

ClawRunr addresses a real gap in AI agent design by making retries, persistence, scheduling, and monitoring first-class through a JobRunr substrate beneath Spring AI's ChatClient.

Explicit premises:

Long-running agent work needs durability against failure and process restarts
JobRunr already provides retry, cron, and dashboard monitoring for background jobs
Integrating JobRunr beneath Spring AI's ChatClient inherits these guarantees for agent tasks

Implicit premises:

Existing agent frameworks treat durable execution as an afterthought
Production-grade agents require the same operational properties as background jobs

Inferences:

Therefore ClawRunr's design choice is structurally novel for the Java ecosystem
Therefore agents built on it should behave more like production workloads than transient chat sessions

The architecture (Spring Modulith + Channel interface + workspace files) yields extensibility without redeployment.

Explicit premises:

base/app/plugin separation isolates core from opt-in capabilities
Channel interface lets new surfaces be added as plugins
AGENT.md, INFO.md, and SKILL.md files are runtime-scanned, not compiled

Implicit premises:

Users prefer convention-driven, file-based configuration over code changes
Hot-loading capabilities is preferable to redeploy cycles

Inferences:

Therefore end users can extend ClawRunr without Java development skills
Therefore ecosystem growth (Slack, WhatsApp, more skills) is mechanically straightforward

Local execution against Ollama makes ClawRunr suitable where data sovereignty matters. strength: high

Explicit premises:

ClawRunr runs on users' own hardware
Ollama is supported as an LLM provider
When Ollama is selected, no data leave the user's hardware

Implicit premises:

Open-weights local models are capable enough to drive useful agent behavior

Inferences:

Therefore regulated or privacy-sensitive deployments can use ClawRunr end-to-end on-prem

Narratives

Evolution from PoC to community project: JavaClaw (a Java port of OpenClaw) was a proof-of-concept demonstrating JobRunr applied to AI agents, then rewritten by the JobRunr team and released under LGPL-3.0 as ClawRunr.
→ Frames the project as a maturation rather than a fresh launch; lends credibility through the JobRunr team's stewardship.
Production-readiness shift: AI agents are moving from chat demos to durable, monitored background workers; ClawRunr is positioned as the Java ecosystem's answer to this shift.
→ Sets the project's contribution within an industry narrative about agents becoming infrastructure-grade workloads.
Java-ecosystem self-sufficiency: Java 25 + Spring Boot/AI/Modulith + JobRunr signals that the Java community can host first-class agent runtimes without leaving its stack.
→ Appeals to Java enterprise audiences who might otherwise feel that agent tooling is Python-only.

Questions addressed

✅ What is ClawRunr and what problem does it solve?
An open-source Java AI agent built on JobRunr that treats retries, persistence, scheduling, and monitoring as first-class concerns for long-running agent work.
✅ What is the technology stack and architecture?
Java 25, Spring Boot, Spring AI, Spring Modulith, JobRunr, H2; three modules (base, app, plugin); Channel abstraction; workspace-file prompt composition.
✅ How are tasks stored, scheduled, retried, and monitored?
Tasks as Markdown files under workspace/tasks/ with explicit lifecycle states; recurring tasks via JobRunr cron with retries and dashboard monitoring.
✅ How are tools and skills added or discovered?
Tools via @Tool annotations and MCP servers (HTTP/stdio); optional Lucene-based dynamic tool discovery; skills via SKILL.md files under workspace/skills/.
✅ What channels and LLM providers are supported?
Web chat, Telegram, Discord; OpenAI, Anthropic, Ollama.
✅ How is privacy and local execution handled?
Ollama selection enables fully local, end-to-end execution with no data egress.
✅ What is the project's provenance and license?
PoC origins as JavaClaw, rewritten by the JobRunr team, released under LGPL-3.0.
✅ What are the known limitations and roadmap?
Playwright requires restart after first install; web chat weaker than Telegram for recurring output; roadmap includes Slack, WhatsApp, memory, planning, more tools and skills.
⭕ How does ClawRunr compare to existing agent frameworks (LangGraph, CrewAI, AutoGen, Semantic Kernel)?
Not addressed; comparative positioning is implicit at best.

Gaps

evaluativeNo empirical comparison to competing agent frameworks (LangGraph, CrewAI, AutoGen, Semantic Kernel) on durability, latency, or cost.
Read next: Hands-on benchmarks or independent reviews of Spring AI vs. LangGraph for durable agent workloads
securityShell execution and broad file access are powerful tools, but sandboxing, permission scoping, and audit trails are not described.
Read next: Security guidance for self-hosted agents with shell/tool access (Anthropic's Computer Use safety, OWASP LLM Top 10)
scalabilityH2 embedded DB suggests single-node use, but limits (task throughput, concurrent channels) are unstated.
Read next: JobRunr Pro / enterprise deployment guidance for scaling to external DBs
conceptualMemory management is only mentioned as a roadmap item; current strategy for long-context or cross-session memory is unclear.
Read next: Memory patterns in Spring AI and broader agent memory literature (MemGPT, LangGraph checkpointing)
closed byThe Production Gap: 5 Patterns for Building Long-Running AI Agents — Provides the agent memory-pattern guidance the gap's reading direction called for: long-term vs working memory as architecturally distinct stores, memory drift as a named failure mode, and policy enforcement at the memory boundary.
observabilityRich monitoring is offered for jobs, but no treatment of LLM-side observability: token accounting, budget controls, prompt regression testing, tool-call correctness, reasoning quality.
Read next: LLM observability tooling (LangSmith, Langfuse, Arize) and evaluation frameworks
designUnclear how SKILL.md files interact with the dynamic tool discovery index, or how skill collisions/overrides are resolved.
Read next: ClawRunr source for skill registration logic; Anthropic skills documentation for collision/precedence patterns
licensingLGPL-3.0 implications for embedding ClawRunr in proprietary products are not analyzed.
Read next: FSF guidance on LGPL-3.0 in enterprise software; comparison with Apache 2.0 for similar projects
evaluationNo methodology for agent quality (success rates, regression testing) accompanies the production-readiness claims.
Read next: Agent evaluation literature: AgentBench, SWE-bench, internal eval harness design

JobRunr Introduces ClawRunr, an Open-Source Java AI Agent

Structure

Key terms

Propositions

Arguments

Narratives

Questions addressed

Gaps

Related documents