MartinLoop

MartinLoop runs around coding agents to keep software jobs within agreed boundaries and check their results. It's for developers and engineering teams who want to delegate longer tasks while retaining limits, recovery information and a record of what happened.

• Job limits: Spending, attempt, time and path limits constrain a run, while stop conditions end work that stalls or keeps retrying.
• Verification: Configured tests or other checks determine whether a run qualifies as complete; the outcome records whether work passed, stopped or needs review.
• Recovery and evidence: Preserved state supports retries, repair, rollback or human handoff, and receipts record costs, check results and stop reasons.
• Agent choice: Native integrations work with Claude Code, Codex and Gemini CLI, while MCP hosts can execute jobs and inspect results without changing the surrounding job rules.
• Inspectable history: Local run records carry signatures that make later edits detectable, and share bundles provide redacted receipts and readable recaps.

The free, open-source core runs on your machine under Apache 2.0 and needs no account to start. The coding agent handles model access, so local execution doesn't mean its model requests stay local. OpenAI-compatible inference workflows can use Ollama, LM Studio, llama.cpp or hosted endpoints, but this path isn't a finished autonomous workspace-editing worker.

A separate paid hosted dashboard provides run analysis; it doesn't run locally. Anonymous usage telemetry is enabled after disclosure and can be disabled. It excludes source code, prompts, model output and receipt contents. Passing verification establishes that the configured checks passed, rather than guaranteeing correct or secure code.