Rewrite without the all-or-nothing weekend
Rewrites fail on sequencing, not on technology: a plan that lives in meetings, dependencies found mid-cutover, one big-bang cutover nobody can roll back.
The Bunny Lab captures real inputs at the seam of the running system, plans the waves from the module graph, translates through a knowledge base and a memo cache, repairs with the compiler as the oracle, and ships nothing without an equivalence verdict.
What you get
- Golden inputs recorded from production, types included, before a line is rewritten.
- A deterministic plan: module graph condensed, cycles broken and recorded, waves ordered leaves-first.
- Each unit translated into memory-safe Rust, checked and repaired against the compiler, verified against the original on the same inputs.
- A receipt per unit and an append-only ledger of every run, cost and gate.
How it works
You pick the workload.
The one the assessment shows is costing the most, with a stable interface.
We capture and plan.
Golden cases from production and a wave plan your team reviews before the first translation.
We translate in waves.
Each unit verified-equivalent or it does not ship; each wave reversible; a shadow deploy before any traffic.
BEFORE YOU START
Before you start
Which languages qualify?
Python and TypeScript today, with SQLAlchemy models handled by a schema pass; Java, C/C++ and COBOL workloads are scoped in the assessment. CPU-heavy code with a stable interface goes first.
We have no tests. Is that a problem?
No. The verification corpus is recorded from production, not from a test plan. Every verdict is judged against what the system actually did.
What happens if a wave fails?
It rolls back. Waves are planned from the module graph and reversible by design. Nothing ships without the equivalence verdict.
Stop paying for the code you are afraid of.
Bring one workload and see it rewritten, verified and running in shadow.