Vinv

What it is

Vinv is a set of tools and algorithms that enables AI agents to test, fix, and optimize Python codebases. It runs Python services with zero code changes and generates real runtime evidence—traces, argument values, failing frames—for AI agents to use when fixing issues. After the agent proposes a fix, Vinv re-runs the code to verify it works, ensuring the fix passes acceptance tests before deployment.

The platform provides multiple detection methods including an HTTP exerciser for endpoints, differential oracles for comparing against reference implementations, fault injection at dependency boundaries, concurrency oracles for detecting deadlocks, environment compatibility checks, golden I/O baselines, dead code detection from actual execution, and runtime analysis for hotspots and memory leaks.

Vinv works as an MCP server compatible with any AI coding agent—Claude Code, Cursor, Windsurf, Codex, Gemini CLI, and others—eliminating the need for new subscriptions or API keys. It runs 100% locally with an embedded model that never uploads data, maintaining complete privacy and control.

Questions from the bench

How does Vinv verify that a fix actually works?

Vinv restarts the app in a fresh process, requires the port to actually serve, and runs acceptance tests that were written before the fix and which the agent never sees. If tests fail, the findings feed the next attempt and one click reverts everything the fix touched.

Does Vinv require code changes or SDK integration?

No. Vinv runs Python services with zero code changes and traces without requiring any SDK integration.

How does Vinv compare with frontier AI models for code fixing?

In testing, a commodity model with Vinv evidence found 4 bugs and 1 optimization, while a frontier model without evidence found only 1 bug, and a commodity model without evidence found nothing. The evidence made the difference, not the model size.

My AI coding agent says done but the tests fail. How does Vinv help?

Vinv never accepts the agent's own claim. When a fix comes back it restarts the app in a fresh process, requires the port to actually serve, and runs acceptance tests generated before the fix that the agent never sees. Failures feed the next attempt; one click reverts everything the fix touched.

What does Auto-Pilot exercise today?

Services and APIs. Auto-Pilot discovers runnable Python services, brings them up under tracelens, and drives their HTTP and related entrypoints. Plain library or in-process function driving is not the Auto-Pilot surface right now.