JobRunr Unveils ClawRunr: An Open-Source Java AI Agent Pioneering On-Premise, Persistent, and Enterprise-Ready Task Automation

JobRunr, a prominent name in the realm of open-source background job processing, has officially launched ClawRunr, an innovative open-source Java AI agent designed to operate directly on users’ own hardware. This introduction marks a significant step forward in addressing long-standing challenges associated with the reliability and persistence of AI agents, particularly concerning scheduled, recurring, and one-off background tasks. Formerly known as JavaClaw, a Java iteration of the OpenClaw project, ClawRunr integrates conversational interaction with robust, persistent task execution, enabling a wide array of automated functions from setting reminders and automating workflows to connecting various tools via the Model Context Protocol (MCP). Users can engage with the agent through multiple channels, including web chat, Telegram, and Discord, signifying a versatile and accessible approach to AI-powered automation.

The core problem ClawRunr aims to solve is a prevalent one in AI agent design: making crucial aspects like retries, persistence, scheduling, and monitoring first-class citizens in the lifecycle of long-running agent work. While many AI agents excel at real-time conversational tasks, their ability to reliably execute and manage complex, multi-step, and time-bound operations has often been limited. JobRunr’s foray into this space leverages its established expertise in dependable background task processing to imbue AI agents with enterprise-grade reliability and manageability.

The Architectural Foundations of ClawRunr: A Deep Dive into its Robust Stack

ClawRunr is meticulously engineered upon a modern and robust technology stack, primarily built on Java 25, reflecting a commitment to leveraging the latest advancements in the Java ecosystem. The agent heavily utilizes components from the Spring ecosystem, including Spring Boot for rapid application development and deployment, Spring AI for seamless integration with large language models (LLMs), and Spring Modulith for creating well-structured, modular applications. The underlying power for task execution and persistence is inherently derived from JobRunr itself, ensuring that all agent-managed tasks benefit from JobRunr’s proven capabilities.

For state persistence, ClawRunr employs an embedded H2 database, offering a lightweight yet effective solution for local data storage. This choice aligns with the project’s emphasis on self-contained, on-premise operation. Furthermore, ClawRunr boasts broad compatibility with leading LLM providers, supporting OpenAI, Anthropic, and crucially, Ollama. The inclusion of Ollama is particularly noteworthy as it facilitates entirely local execution of the agent, ensuring that no data leaves the user’s hardware when a local model is chosen. This capability addresses critical concerns around data privacy, security, and compliance, making ClawRunr an attractive option for organizations and individuals with strict data governance requirements.

Architecturally, ClawRunr adheres to a modular design philosophy, leveraging Spring Modulith to segment its functionalities into three distinct modules. The base module forms the core of the agent, encompassing fundamental elements such as the agent’s intelligence, task management, tool integration, communication channels, and configuration. The app module serves as the Spring Boot entry point, providing the initial onboarding user interface and a built-in web chat for immediate interaction. Finally, the plugin module offers opt-in capabilities, allowing users to extend the agent’s functionality with integrations like Brave Search, Discord, Telegram, and advanced browser automation via Playwright.

The agent’s core functionality wraps Spring AI’s ChatClient, providing a unified interface for interacting with various LLMs. Tools, which are essential for an AI agent to perform actions, are registered through simple @Tool annotations, making it straightforward for developers to extend the agent’s capabilities. A flexible Channel interface decouples message sources from the agent’s processing logic, allowing for easy integration of new communication platforms. This interface is implemented by the built-in web chat in the app module and by the Telegram and Discord plugins. The system prompt, which guides the agent’s behavior and context, is dynamically composed at runtime from two workspace files: AGENT.md for core instructions and INFO.md for environmental context, allowing for highly customizable and adaptive agent personas.

JobRunr’s Role: Empowering Robust Task Execution

The integration of JobRunr is central to ClawRunr’s value proposition, directly addressing the aforementioned challenges of persistence, scheduling, and monitoring in AI agents. JobRunr handles all asynchronous task execution for ClawRunr, transforming ephemeral AI instructions into reliable, long-running processes. User-created tasks are stored as Markdown files within the workspace/tasks/ directory, organized by date, and tagged with a state in a defined lifecycle: todo, in_progress, completed, or awaiting_human_input. This file-based persistence ensures that tasks are not lost, even if the agent or system restarts.

For recurring tasks, ClawRunr utilizes JobRunr’s robust cron support. These tasks are persisted under workspace/tasks/recurring/ and are scheduled to run at specified intervals. A key benefit of this integration is JobRunr’s automatic retry mechanism, which ensures that tasks failing due to transient issues are automatically re-attempted, significantly enhancing reliability. Furthermore, JobRunr’s comprehensive dashboard provides real-time monitoring of all tasks, offering visibility into their status, execution history, and any potential failures. This level of operational insight is critical for enterprise environments where accountability and observability are paramount.

This integration builds upon JobRunr’s established track record. InfoQ, for instance, previously highlighted JobRunr 8.0, which introduced features like carbon-aware job processing – allowing jobs to be scheduled based on the availability of green energy – Kotlin serialization support, and enhanced dashboard notifications. These advancements underscore JobRunr’s continuous innovation in the background processing space, a foundation that now directly benefits ClawRunr.

Expanding Capabilities: Tools, Skills, and Seamless Integration

ClawRunr ships with a comprehensive suite of built-in tools designed to perform a wide range of tasks. These include essential functionalities for task management, executing shell commands, accessing files, and performing web scraping. The agent also integrates an MCP server over both streamable HTTP and stdio transports, facilitating interoperability with other tools and services that adhere to the Model Context Protocol. For enhanced search capabilities, ClawRunr supports Brave Search when configured, providing access to a privacy-focused search engine.

A significant optional feature is the Playwright integration, which enables advanced browser automation. This allows ClawRunr to perform complex web interactions such as page navigation, clicking elements, executing JavaScript, and taking screenshots, effectively giving the AI agent a "browser" to interact with the web. This opens up possibilities for sophisticated web-based automation scenarios. To manage the growing number of tools efficiently, ClawRunr exposes an opt-in dynamic tool discovery mode. Based on Spring AI’s Tool Search Tool pattern, this feature uses Lucene keyword search to surface only the most relevant tools at request time, rather than including all tool definitions in every LLM prompt. This optimizes prompt length, reduces token usage, and improves the efficiency and focus of the AI agent.

Beyond predefined tools, ClawRunr features an innovative file-based skill system, allowing users to easily add new capabilities without requiring code changes or redeployments. Users simply create a new directory under workspace/skills/ and place a SKILL.md file within it. The runtime scans this directory and makes the instructions within these Markdown files available to the agent, providing a highly flexible and extensible mechanism for customizing and expanding the agent’s knowledge and abilities. This approach democratizes the process of enhancing AI agent capabilities, allowing non-developers to contribute to the agent’s skill set.

workspace/
├── skills/
│   ├── my-skill/
│   │   └── SKILL.md

User Experience and Deployment: Onboarding and Configuration

ClawRunr prioritizes a streamlined user experience, particularly during the initial setup. A guided onboarding flow walks users through the essential configuration steps. This includes selecting an LLM provider (OpenAI, Anthropic, or Ollama), configuring credentials and model parameters, customizing the agent’s system prompt to align with specific needs, optionally setting up the MCP server, and configuring communication channels for Telegram or Discord. This step-by-step process ensures that users, even those new to AI agents, can get ClawRunr up and running with minimal friction.

Configuration settings are persistently stored in the standard application.yaml file, where channels are declaratively wired up. For example:

agent:
  channels:
    telegram:
      token: <your-bot-token>
      username: <your-telegram-username>
    discord:
      token: <your-discord-bot-token>
      allowed-user: <your-discord-user-id>

Changes applied to this configuration are reflected immediately, enhancing the agility of deployment and management. A critical aspect of ClawRunr’s design is its ability to operate entirely offline when Ollama is selected as the LLM provider. In this scenario, ClawRunr runs end-to-end against a local model, ensuring that no data leaves the user’s hardware. This feature is a powerful differentiator for users concerned about data privacy, regulatory compliance, or operating in environments with limited or no internet connectivity.

The Journey from Concept to Open-Source Reality

ClawRunr’s genesis can be traced back to a proof of concept initiated by the JobRunr team. The primary motivation was to demonstrate the applicability and power of JobRunr’s robust background processing capabilities within the emerging domain of AI agents. The team recognized that while AI agents offered unprecedented intelligence, their operational reliability and persistence for long-running tasks were often lacking. This gap presented a unique opportunity to combine JobRunr’s strengths with AI.

Following the successful proof of concept, the JobRunr team undertook a comprehensive rewrite of the implementation, refining the architecture, enhancing features, and ensuring stability. This dedication culminated in the release of ClawRunr to the broader Java community under the LGPL-3.0 license. This open-source licensing encourages widespread adoption, collaboration, and further development by the community, aligning with JobRunr’s commitment to open-source principles. The transition from JavaClaw, an earlier Java variant of OpenClaw, to the more robust and feature-rich ClawRunr, reflects a maturation of the project and a clearer vision for its role in the AI agent landscape.

Strategic Implications: Shaping the Future of Enterprise AI Agents

ClawRunr’s introduction carries significant implications for the future of AI agents, particularly within enterprise and privacy-conscious environments.

On-Premise AI and Data Sovereignty: The ability to run AI agents on users’ own hardware, especially with local LLMs like Ollama, is a game-changer for data sovereignty and privacy. In an era of increasing data regulations (e.g., GDPR, CCPA) and heightened concerns about cloud data security, ClawRunr offers a compelling solution. Enterprises handling sensitive customer data, proprietary information, or operating in highly regulated industries can deploy AI agents without the inherent risks of sending data to external cloud-based LLM providers. This model empowers organizations to maintain full control over their data lifecycle, reducing compliance burdens and mitigating security vulnerabilities associated with third-party data processing.

Reliability and Enterprise Adoption: The foundation of JobRunr directly addresses one of the biggest hurdles for enterprise adoption of AI agents: reliability. Traditional AI agent demonstrations often focus on impressive conversational abilities but gloss over the complexities of ensuring tasks complete, retrying failures, scheduling recurring processes, and monitoring performance at scale. ClawRunr, by integrating JobRunr, brings enterprise-grade reliability, observability, and task management to AI agents. This makes it a much more viable and trustworthy solution for critical business processes, from automated customer support workflows to internal operational tasks, where failures or inconsistencies are unacceptable.

Empowering Java Developers: For the vast Java developer community, ClawRunr provides a robust, familiar, and extensible framework for building and deploying AI agents. By leveraging well-established technologies like Java, Spring Boot, and the Spring AI ecosystem, it lowers the barrier to entry for Java developers looking to integrate AI capabilities into their applications. This move could catalyze a surge in Java-based AI agent development, fostering innovation and creating a rich ecosystem of tools and skills.

Democratizing AI Agent Development: The open-source nature and the file-based skill system democratize the process of enhancing AI agent capabilities. Developers and even non-technical users can contribute to the agent’s skill set without complex code changes or recompilation. This flexibility encourages experimentation and community-driven development, accelerating the evolution of ClawRunr and its potential applications.

The Model Context Protocol (MCP): ClawRunr’s integration with MCP is forward-thinking. MCP aims to standardize how AI agents interact with tools and services, creating a more interoperable and less fragmented AI ecosystem. By supporting MCP, ClawRunr positions itself as a potential hub for various AI-powered automations, facilitating seamless communication and collaboration between different agents and tools.

Forward Momentum: Known Issues and Future Roadmap

The JobRunr team has been transparent about known issues and their future roadmap for ClawRunr. Currently, they acknowledge that Playwright integration requires a restart after its initial browser installation, which can be an inconvenience. Additionally, recurring task output is presently better supported through the Telegram channel than through the built-in web chat interface, indicating an area for improvement in the user experience across different communication channels.

The roadmap for ClawRunr is ambitious and reflects a commitment to continuous enhancement. Planned additions include support for more communication channels such as Slack and WhatsApp, which would significantly broaden the agent’s reach and utility in various professional and personal contexts. Further improvements are slated for memory management, enabling the agent to maintain context and learn more effectively over longer interactions. Enhancements in task planning capabilities will allow the agent to break down complex goals into more manageable sub-tasks. The team also plans to introduce a wider array of built-in tools and expand the skill system, further increasing ClawRunr’s versatility and power. These planned features indicate a strategic direction towards making ClawRunr an even more intelligent, versatile, and user-friendly AI agent.

Conclusion: A New Paradigm for AI-Powered Automation

ClawRunr represents a pivotal development in the landscape of AI agents. By merging the conversational intelligence of modern AI with the unwavering reliability and structured task management of JobRunr, it offers a robust, persistent, and enterprise-ready solution for automated workflows. Its emphasis on on-premise execution, modular architecture, and open-source licensing positions it as a significant player in addressing the practical challenges of deploying AI agents in real-world scenarios. For Java developers, enterprises prioritizing data privacy, and anyone seeking a dependable AI agent for scheduled and recurring tasks, ClawRunr presents a compelling and powerful new paradigm for AI-powered automation.

More comprehensive details about ClawRunr, including its technical specifications and usage guides, can be found on the project’s official website at clawrunr.io and its GitHub repository at github.com/jobrunr/JavaClaw.