TestPilot turns a Python repository diff into candidate regression tests. It identifies changed functions, asks a planner and editor for tests, runs those tests in a temporary subprocess with a timeout, and repairs failures for bounded rounds. A code-bug verdict preserves a failing test instead of forcing it green.
The Python prototype supports a Nebius Token Factory backend for NVIDIA Nemotron; its documented evaluation uses a deterministic scripted model. The pipeline produces a patch, coverage information and a token ledger. Live project-specific inference and model quality remain to be independently evaluated.
Generated tests need human review. The execution environment uses temporary copies and timeouts; it is not a complete security boundary for untrusted code.
Log in or sign up for Devpost to join the conversation.