TestMind AI

TestMind AI is a test-quality tool that proves, with measured data rather than estimates, whether a python repository's test suite would actually catch a real bug and then uses AI to write the tests that are missing. Coverage only tells you which lines ran; it says nothing about whether your tests would notice a defect. TestMind AI injects small bugs into the code on purpose (flipped comparisons, swapped operators, changed constants, removed exceptions) using its own AST based mutation engine, no mutmut or stryker, then reruns the suite and reports how many of those injected bugs survive undetected. Beyond mutation testing, the deterministic engine also measures line coverage, ranks files by risk, discovers FastAPI endpoints with no test coverage and smoke-tests them, and runs Playwright-based visual regression with pixel diffing and basic accessibility checks. Every number in its dashboard comes from an actual measurement , nothing is estimated, rounded, or extrapolated by a model. On top of that measurement layer sits an AI fix loop: two agents running in sequence, a test writer and a critic, powered by either IBM watsonx.ai or Google Vertex AI (switchable per call). They write new tests aimed at the surviving mutants through a single tool that is hard-guarded to only write inside tests it can never modify source code. A suite of integrity guardrails, verified in CI, specifically neutralizes ways an AI could game its own score: reading secrets out of the environment, writing assertion-free tests, or artificially inflating the mutation score. A parallel, one-lane-per-file agent swarm is being built as a faster alternative to the sequential loop, but is not yet wired into the default path. The whole engine is exposed through a CLI, a web dashboard, and an MCP server, so any MCP-capable AI agent can drive the same measurements directly