a test-driven bug-fixing agent

Every fix ships with
the test that proves it.

AI tools hand you a patch and ask you to trust it. tdfix won't propose a fix it can't prove — it writes a test that reproduces the bug, then patches until that test goes from red to green, with your whole suite still passing.

✓ pay once, no subscription ✓ bring your own key or run local ✓ your code never leaves your machine
the problem

A patch you can't verify is just a guess
with good grammar.

The bottleneck was never generating code — it's trusting it. tdfix is built around a verdict, not a vibe.

✗ Patch-only AI

Spits out a diff. You read it, hope it's right, and have no test guarding the bug from coming back. Confidently wrong fixes slip through review.

✓ Test-driven repair

Generates a failing test that reproduces the bug first, then proves the fix flips it green — and leaves that test behind forever as a regression guard.

how it works

Reproduce. Prove. Fix. Verify.

A self-verifying loop with a built-in oracle — so it can run at your terminal or unattended in CI without handing you false positives.

Reproduce

From an issue, stack trace, or a sentence, it writes a test that fails because of the bug.

Prove the fail

Confirms the test fails for the right reason, with a flakiness check — not an import error in disguise.

Fix

Iterates a patch until the frozen test passes. The patch can't touch the test — no cheating.

Verify

Red→green proven, full suite green, and reverting the patch re-breaks the test. Then it opens a PR.

Runs locally on Ollama, or BYOK (Anthropic / OpenAI) Drops into CI as a reproduce-and-fix bot One-time purchase. Lifetime updates.
shape v1

How would you
use it?

v1 is being built in the open. Tell me the bug you'd point it at first, the language you'd want next, or the CI flow you wish it had. The best ideas go straight into the roadmap — and early responders get launch access.