Binary Bedrockbinarybedrock.com
Closed beta

Every 1 counts.

Every 0, too. Binary Bedrock mathematically proves that changed code — an AI refactor, a compiler trick, a hand optimization — computes exactly what the original computed. Bit for bit, on all inputs. Not tested. Proven.

runs on-prem & air-gapped · your source never leaves your machine · zero model calls in the prover

001What it does

Three verdicts. No opinions.

Give it two versions of a function — before and after any change. It compiles both and asks an SMT solver whether they are equivalent on every input. The answer is one of three, and it is never a guess.

✓ PROVEN exit 0

The change is behavior-preserving — a machine-checked equivalence certificate is produced. Safe to merge. This is a mathematical proof over all inputs, not a test suite's sample.

✗ DIVERGES exit 1

The change alters behavior — and you get the counterexample: a concrete input where old and new disagree. The bug report writes itself.

? CANNOT CERTIFY exit 2

The proof exceeded its budget. The gate stays closed — an unproven change is not treated as a safe change. Binary Bedrock fails safe, never silent.

010How to use it

Two ways in. Both keep your code yours.

AI assistants write more of your code every week. Binary Bedrock is the gate they pass through — from your editor for single functions, or on your own hardware for whole trees.

In your editor beta · hosted demo

  1. Add one MCP server to Claude Code, Cursor, or any MCP-capable agent — paste the config below with your beta token.
  2. Ask your assistant to verify its own work: “refactor this, then prove it with Binary Bedrock.” The agent calls verify_refactor and shows you the verdict.
  3. Merge on PROVEN. On DIVERGES you get the input that breaks it — before it ships, not after.
{
  "mcpServers": {
    "binary-bedrock": {
      "url": "https://mcp.binarybedrock.com",
      "headers": { "Authorization": "Bearer <your-token>" }
    }
  }
}

The demo endpoint verifies code you paste — it holds no keys, calls no models, and never executes your code. For anything sensitive, use the on-prem kit.

On your metal on-prem kit · air-gapped

  1. Load one container image — we hand you a checksummed bundle; docker load, no registry, no account, no network required.
  2. Point it at your code: bb ingest takes a source tree or your build's compile_commands.json and runs every function through the proof funnel automatically.
  3. Gate your CI: bb verify-diff old/ new/ proves every changed function across whole directory trees — one exit code decides the merge.
  4. Read the funnel: every function that couldn't be verified is listed with a named reason and what would make it eligible. No silent drops.
$ sha256sum -c CHECKSUMS.txt   # verify the bundle
$ docker load < bb-toolchain.tar.gz
$ ./bb ingest compile_commands.json
$ ./bb verify-diff before/ after/
011Numbers

Every number below is from a recorded run.

We publish what the tool measured — including what it can't do yet. That discipline is the product.

0 / 17,554
false “PROVEN” verdicts across 17,554 adversarial attack attempts on the firewall — deliberately broken code planted to fool it. The zero has never broken.
3.7%
of a frontier AI model's ordinary refactors silently changed behavior in our recorded field studies (126 attempts, pooled). Every single one was caught. None were blessed.
98.2%
of valid AI rewrites classified decisively — PROVEN or DIVERGES with a counterexample — rather than refused. The gate answers; it rarely shrugs.
42 proven
functions on real, unseen open-source code (zlib + littlefs, 219 functions, nobody hand-picked) — 61% of everything that entered the funnel. Zero wrongly accepted.
59% smaller
a proven-identical routine on Cortex-M4: 34 → 14 bytes, formally verified equivalent while shrinking. Verified optimization, not vibes.
seconds
typical proof time for a single function — median well under a minute; hard cases are given a budget and refused honestly when it runs out.

methodology & full runs available to beta partners · verdicts anchored in industry-standard formal tools (translation validation + SMT), independent of any AI model

110Who it's for

Plain outcomes. No binary talk.

Three verbs, every industry: reduce size, speed up, prove it didn't break.

AI & ML

Let AI write code — ship it only when it's proven right. And shrink the math inside on-device AI so it runs faster and cheaper.

Defence & aerospace

Prove that every change to critical code changes nothing it shouldn't. Works fully offline, inside your walls.

Crypto & security

The routines that checksums, signatures and money depend on — proven identical after every optimization.

Embedded

Smaller firmware, longer battery, cheaper hardware — with a proof that behavior didn't change.

Chips & silicon

Squeeze more out of every core you ship — verified separately for each target you sell.

Your team

One config line in your editor, one gate in your CI. Merge on PROVEN; get the exact breaking input otherwise.

Why big code doesn't scare us
pieces of code proven together (recorded run, up to 64) time to prove ✗ all-at-once: gives up at 2 ✓ Bedrock: split, prove pieces, stay linear

Provers choke on big functions. Bedrock automatically splits code at safe boundaries, proves each piece, and proves the splitting itself — so proof time grows in a straight line instead of hitting a wall.

100What it does not do

Said out loud, on purpose.

  • No “fault-free software” claims. Binary Bedrock proves a change preserves behavior — it does not prove your original code was right.
  • Floating point rewrites rarely prove bit-exact — that's the honest mathematics of FP, and we say so instead of hand-waving.
  • Unbounded loops and recursion end in CANNOT-CERTIFY — the gate stays closed rather than guessing.
  • C and C++ carry the product today. Rust and GPU-kernel coverage is experimental and labeled as such.
101Closed beta

Prove your next merge.

Request access below, or create an account directly — once approved you create your own MCP tokens in the portal. We are onboarding a small number of teams — embedded & firmware shops, and engineering orgs using AI assistants heavily. You get a token for the editor demo, the on-prem kit, and a direct line to us.