Debugging code you did not write
Trace a fault through unfamiliar code, with the evidence for each likely cause.
The problem
Engineers often spend long periods tracing a fault through unfamiliar code. When an AI tool suggests a fix, it often arrives without sources, so there is no quick way to check it.
What works today
In our internal prototype, inside our team only. It is not available to customers yet.
- Root-cause tracing follows a fault across logs, code, tests and recent changes.
- It returns a ranked list of likely causes, with the evidence for each.
- It works on TypeScript and Python code.
What is planned
Targets, counted from the plan's first month, February 2027. They may move.
- Root-cause tracing with the first three design partnersQ2 2027
- A published time-to-diagnosis benchmark with partnersQ3 2027
- Repository integrations, with team plansQ4 2027
- IDE integrationsBy Q3 2028
Example: a checkout that times out
Invented example. The shop, the files and the numbers are made up.
An invented online shop sees checkout time out under load. Belay reads 14 files, 6 tests and the last 20 commits. It returns three likely causes in order of evidence: retries that hold connections open in payments/client.ts, a query per item in cart/totals.ts, and a missing index in an orders migration. No files are changed. The engineer decides what to do next.
We are looking for design partners
5 to 8 engineering teams that work on large or legacy codebases.