A developer has released RunOnMine, an open-source project that implements local policy and approval workflows for AI agent access to your machine. The tool surfaced on Hacker News on August 9th as a Show HN submission, offering developers a framework to control what AI systems can read, write, and execute on their local environments.

Why This Matters for AI Development

As AI agents become more capable of executing code, reading files, and performing system operations, the question of granular permissions becomes critical. Current approaches often rely on broad access grants or manual oversight. RunOnMine proposes a policy-based layer that lets developers define explicit rules about what actions AI systems can take, with approval workflows for sensitive operations.

How Policy-Based Access Control Works

The core idea behind RunOnMine is straightforward: instead of giving an AI agent unchecked access to your filesystem and system resources, you define policies that govern what the agent can do. These policies can require approval for certain operations—file deletions, network connections, execution of specific binaries—while allowing routine tasks to proceed automatically.

Defining Policies in Practice

Policies in RunOnMine are defined declaratively, specifying which operations require explicit user approval versus which can run autonomously. For example, a developer might configure a policy that allows an AI agent to read files within a designated workspace directory without prompts, but requires user confirmation before executing shell commands or writing outside project boundaries. The system supports wildcard patterns for path matching and can differentiate between read, write, and execute permissions at a granular level.

The Approval Workflow in Action

When an AI agent attempts an action that triggers a policy rule requiring approval, RunOnMine pauses execution and presents the user with context about what operation is being requested—including the target file or command, its arguments, and which policy triggered the hold. Users can approve, deny, or add exceptions for future similar operations. This creates an audit trail of AI actions and prevents silent credential escalation.

Real-World Use Cases

For developers building AI-powered tooling, RunOnMine offers a way to ship products with appropriate guardrails baked in. Consider an IDE plugin that uses an LLM to refactor code: the policy layer can ensure the agent only modifies files within the current project scope, never touches configuration files outside of explicit allowlists, and prompts before running any shell commands like git operations or package installations.

The Developer Experience

Developers integrating RunOnMine into their applications get a clean API for defining policies and handling approval callbacks. The framework is designed to be embedded without significant overhead, making it practical for both experimental AI tools and production-grade systems where permission boundaries matter for security and compliance.

Looking Ahead

The project is early-stage, reflected in its modest Hacker News score of 2 points at publication time. The source code is available on GitHub for developers interested in exploring policy-based AI access controls or contributing to the project's development.

Key Takeaways

  • RunOnMine implements local policy enforcement for AI agent operations
  • Designed for developers building tools that need granular permissions control
  • Open-source and available now on GitHub
  • Addresses an emerging pain point as AI agents become more capable

The Bottom Line

This is exactly the kind of infrastructure tooling the ecosystem needs right now. As we hand more capability to AI systems, having proper permission boundaries isn't optional—it's essential for building trustworthy applications.