Martin Loop.

Hard budget stops

Set a dollar ceiling with --budget. MartinLoop tracks spend in real time and stops the run before it overshoots — not after the bill arrives. This feature ensures that users maintain control over their budget and avoid unexpected costs.

12-class failure taxonomy

Every error gets routed to the right fix, such as syntax errors leading to constraint repairs and hallucinations triggering grounding checks. This structured approach eliminates blind retries and enhances the efficiency of error handling.

Evidence-gated completions

A run is only marked done when your --verify command passes, ensuring that no task is considered complete without proof of its success. This feature adds a layer of accountability to the completion process.

Stagnation detection

Loops that stop making progress are exited automatically, preventing agents from running indefinitely without producing results. This feature helps optimize resource usage and maintain efficiency.

Thinking-token tracking

Every token is counted, including reasoning tokens that older tools miss. This feature provides users with a clear view of their spending, ensuring transparency in costs associated with AI operations.

Sub-agent cost rollup

Spend from spawned sub-agents rolls up to the parent session, ensuring accurate attribution of costs at any depth. This feature simplifies financial tracking across complex agent interactions.

Inspectable audit trail

Every action, decision, and approval is captured in a structured record that can be replayed end to end. This feature provides a comprehensive audit trail for compliance and accountability.

JSONL run records

Every iteration is written to disk for replay, allowing users to walk through any run, reconstruct decisions, and provide a finance-ready receipt to the CFO. This feature enhances transparency and accountability in AI operations.

The MartinLoop Dashboard

Real-time loop telemetry, cost-per-task heatmaps, and inspectable run records are provided through the dashboard. This feature allows users to visualize and analyze their AI operations effectively.

Job queue management for parallel agent runs

HeadlessOS manages job queues for multiple agent runs simultaneously, optimizing resource allocation and ensuring efficient processing. This feature is essential for teams running numerous agents at scale.

Approval workflow enforcement

Sensitive changes require sign-off before they run, integrating approval processes into the governance layer. This feature ensures that all significant actions are vetted, enhancing security and compliance.

GitHub integration

This feature allows for seamless integration with GitHub, enabling teams to manage their AI coding agents within their existing workflows. It enhances collaboration and streamlines project management.

Slack alerts on overruns

Receive notifications in Slack when budget overruns occur, ensuring that teams are immediately aware of any financial issues. This feature promotes proactive management of AI operations.