FMToolkit
Unverified AI Code Is a Liability
Prove it before you ship it.
For engineering teams shipping AI-written, AI-refactored, or AI-modernized code who can't afford to find out in production that something broke.
FMToolkit proves, mathematically, whether your code still behaves the way it's supposed to, before you ship it.
AI is now writing, refactoring, optimizing, and modernizing code faster than any team can manually review it — new features, cleanup of existing systems, performance rewrites, or migrating legacy systems. The tools are fast. Knowing whether the result still does what it's supposed to do is the part every team is currently answering with hope, spot-testing, and eyeball review, not proof.
FMToolkit closes that gap. Your existing tests check the inputs someone thought to write down. FMToolkit checks all of them — comparing your code before and after AI touches it and proving, mathematically wherever possible, whether the two versions behave the same way for every input. When they don't match, FMToolkit shows you exactly where and how they diverge, so your team decides whether that's a bug or an intended change, before it reaches customers.
Catch what tests miss
Tests check what you thought to try. FMToolkit checks every input mathematically possible.
Ship with proof, not hope
Move from “we think it works” to “we proved it works” — with a certificate to show for it.
Move at AI speed, safely
Modernize, refactor, and ship faster without trading away correctness to get there.
How FMToolkit Works
- Never modifies your code: only detects behavioral differences and reports what changed.
- Doesn't guess your intent: every finding includes a proof or a concrete counter-example so your team makes the call.
- Fits where you already work: your IDE (via MCP, for Claude Code, Cursor, and similar tools), GitHub pull requests, or any CI/CD pipeline — GitLab, Jenkins, CircleCI, and more.
- Every result comes with a shareable, auditor-readable record: exactly what compliance teams ask for when they ask, “how do you know it still works?”
- Performs best on well-structured, typed code. The included coverage audit shows you exactly which functions are ready today and what changes are needed to bring more into scope.
Every Check Returns One of Four Clear Answers
Proved equivalent
Mathematical certainty — identical behavior for every possible input, not just the ones tested.
High confidence
Extensively tested against edge cases with no divergence found. Strong evidence, short of a full proof.
Found a difference
A concrete, reproducible example of exactly where behavior changed. Your team decides: bug, or intended?
No verdict yet
The code needs a small, specific change to bring it into scope — FMToolkit tells you exactly what.
Who Needs This?
If your team ships AI-generated or AI-modified code into production, and especially if a single regression means a regulatory event, a go-live failure, or a safety incident (banking, insurance, defense, aerospace, telecom, government modernization) — you need proof, not assumptions, that the code still works. Every finding comes with an auditor-readable certificate, so compliance sign-off doesn't mean starting from scratch.
Specification-Based Verification
Most of what's above compares two versions of your code. For teams that want more, FMToolkit can also verify code directly against a written specification — proving a correctness property holds for every input, independent of implementation. Available today as a scoped engagement with our team; ask us to build it out for your codebase.
Learn More or Schedule a Conversation
See FMToolkit on your own code.
Learn More