We got local models to triage the OpenClaw repo for FREE!*

Researchers used local open-source models running on existing hardware to automatically categorize and triage issues and pull requests in a large open-source repository, achieving near-instantaneous notifications at only the cost of electricity. This approach matters because it demonstrates that organizations can reduce dependence on closed commercial models that may be removed or become unavailable, and can instead build their AI infrastructure using models they control and run locally. The system uses an agent framework that allows the local models to read repository files through a restricted shell interface to better understand context before assigning labels to incoming issues and pull requests. The setup processes incoming items through a local database and worker system that prepares relevant context for the model, avoiding the need for the model to directly access external services.