formal verification, applied

Mathematical proof, put to work on the software your industry runs on. 

Banks, hospitals and safety-critical teams now ship code written by AI agents. Predictable Machines proves that code does what your requirements say, on every change. The product is Predictable Code.

Every requirement is traced through the code to a proof. Where the code deviates, the proof shows the hole.

Your rules, in your words. Proven on every change.

A requirement is one sentence your team already says out loud. We turn it into a formal specification that a program checks deterministically. The examples below are illustrative.

Financial services

∎ proven

“A transfer never leaves an account below zero.”

Limits, entitlements, reconciliation. Evidence ready for the audit before the audit is scheduled.

Healthcare

∎ proven

“A record is read only by a member of the care team.”

Access rules, clinical workflows, data handling. The same check for a contractor's code and an agent's.

Safety-critical systems

∎ proven

“No command is sent while the interlock is open.”

State machines, interlocks, failure modes. Every path through the code, not the ones a test happened to take.

Aerospace

∎ proven

“Every telemetry frame is validated before it is used.”

Parsers, protocols, units. A defect is reported with the line that causes it.

Also used in scientific research and streaming media.

One idea from mathematics. Three things it changes on the floor.

for all inputs

Nothing is left to sampling.

Every change is checked against every requirement, plus the defects no spec mentions, like a crash on a missing value.

same premises, same conclusion

The verdict is deterministic.

A program decides what passes, not a model. The same code and spec give the same answer in the editor, in CI and a year later. That determinism is what turns a check into evidence.

counterexample

Each defect arrives named.

The developer knows before pushing, on the line that causes it. A fix counts once it has been checked again.

the product

Predictable Code

Agents write the code. We verify it.

  • Checks code from Claude Code, Cursor, GitHub Copilot, Codex or a person.
  • Runs in VS Code, JetBrains IDEs, the GitHub App, the CLI and agent sessions.
  • Runs in your own GitHub Actions, with your own model key.
  • Free in private beta.

Our articles

Read all

May 6, 2026 · Engineering

AI code pain points, part 4: the merge button and the weight it carries

AI writes the code. AI reviews the code. A human clicks merge. Who is actually responsible? We explore the accountability gap in AI-generated code, why it matters most in regulated industries, and how cryptographically verifiable proof chains change the equation.

Apr 14, 2026 · Engineering

AI code pain points, part 1: the team effect

AI coding tools lose context between sessions and across teams. We explore how Predictable Code addresses the fragmentation problem by turning your codebase into shared, AI-readable specifications.

Mathematicians who ship. Engineers who prove.

Mathematicians and software verification engineers in Seattle and Cádiz, built around one concern: AI is moving faster than the verification needed to scale it responsibly.