All product

Local Capture

The recorder itself

Run one command, pick a folder, and the recording starts. A small daemon reads the transcripts your coding tools already write and stores prompts, responses and diffs in SQLite on your own machine. There is no API key to paste, no provider to choose and no proxy between you and your model.
Book a demo

One command

npx promptwake init, then nothing.

Local by default

SQLite on your own disk.

Under one percent CPU

It reads files rather than intercepting traffic.

Nothing in the path of your work

A recorder that slows people down gets switched off. This one sits beside your tools rather than in front of them: it reads what they already write to disk, which means no added latency on a prompt and nothing to reconfigure when you change model.
Book a demo

Set up once

Point the CLI at a folder. The daemon starts, backfills the history your tools have already written and then watches for new activity. It survives reboots.

  • No API keys
  • No provider selection
  • No configuration file to maintain

Stored where you can see it

Records live in a SQLite database at ~/.promptwake. You can open the file, back it up, or delete it. Cloud sync exists on paid plans and is off until you switch it on, per project.

Secrets are never written

API keys, tokens and credentials are detected and dropped before anything reaches disk. They are not redacted afterwards, so the history cannot leak what it never held.

Works without git

Because capture is independent of commits, work in progress, experiments and branches that were never merged are recorded too — which is usually where the interesting history lives.

What it costs you

No added latency

Your prompts go straight to your model. Nothing is proxied or rewritten.

Under one percent CPU

The daemon idles quietly; it is reading files, not analysing traffic.

No account required

Local capture works without signing in. An account only matters if you want cloud sync.

Setup questions

Which tools does it read?

Twenty-four are recognised automatically, including Claude Code, Cursor, GitHub Copilot, OpenAI Codex, Windsurf, Zed, JetBrains AI and VS Code. Detection is based on the transcript each tool writes, so switching between them mid-project keeps one continuous record.

What if a developer never installs it?

Then that machine produces no record, and the reports say so. Coverage is reported as its own figure rather than assumed, because an unmonitored laptop is a gap and not a clean result.

Can we run it for a whole team?

Yes. Each machine records locally; paid plans sync into a shared workspace with role-based access, and self-hosting keeps everything inside your own infrastructure.

Install it on one repository and look at the history.

Most people are surprised by how much of the last month they cannot otherwise reconstruct.
Book a demoIntegrations

On AI accountability