Aircraft have a black box.
Your AI agents have
nothingno recordno accountabilitya gap in your audita liability you ownPromptWake

When an aircraft is investigated, the recorder answers what happened, when and why. When an AI agent writes code that ends up in production, almost nobody can answer the same question. PromptWake records it — locally, line by line, in a form an auditor can read.
See what gets recorded
24+

AI coding tools recognised automatically

0

API keys, proxies or model routing

<1%

CPU while the daemon idles

1

Command to start recording

100%

Local by default — SQLite on your disk

What the recorder holds

Which agent ran, which model answered, which file it touched, what it changed and the prompt that drove the decision — captured as it happens and kept together. A chat log tells you what was said. This tells you what changed, and lets you prove it months later.
Code provenance

The accountability gap, closed

Your company carries the liability for code it cannot account for. These are the parts that turn AI activity into a record you can stand behind.

Audit Trail

Code Provenance

Compliance Reports

Policy Engine

Local Capture

Every agent on the flight deck.
One recorder, whichever tool is flying.

Evidence, not estimates

Down to the line

A line counts as AI-written only if the text a model produced is still there. Code that was pasted and then rewritten by hand doesn't count — which is why an honest number is usually far lower than a raw one.

Unknown stays unknown

Lines that predate the recorder, or that cannot be matched with confidence, are reported as unknown. They are never folded into the human column to make a chart look tidier.

Yours alone

The recorder writes to SQLite on your own disk. No proxy sits in front of your model, no API key is pasted, and nothing is uploaded unless you deliberately turn it on.

On AI accountability