Synthetic sample / These findings are fictional / Not a report about you
Experiment / The profile between the prompts

How much does
your AI know?

A single question seems harmless. A history of questions can reveal a pattern. Explore a fictional example of what that pattern can become.

Start here / About one minute to explore

See the pattern.

Walk through a synthetic set of conversations, then inspect the fictional report. This does not scan your computer or your accounts.

For your own data

Run it locally.

Read the setup and scope first. Real analysis runs through the separate local tool on your machine.

Read the local setup guide ↓

What happens next

Understand every step.

  1. 01

    Load the sample

    A fixed fictional dataset makes the demonstration repeatable.

  2. 02

    Connect the signals

    A short sequence shows how observations become inferences.

  3. 03

    Inspect the report

    Read the source, the interpretation and the uncertainty separately.

For your own data

Local analysis starts with your choice.

Setup and data scope

The browser demo cannot read your conversation history. The separate command-line tool reads local agent histories when you start it from this website’s source checkout.

Its default sources are ~/.codex/sessions, ~/.claude/projects and ~/.pi/agent/sessions. Starting the command begins the scan. It discards assistant messages and redacts secret-like values from the report.

npm run profiler:local

The tool serves the report on http://127.0.0.1:3219. It can enrich the result using LM Studio on your machine. To run only the deterministic detectors:

npm run build && node tools/local-profiler/server.mjs --no-llm

Open the local server’s report page and choose local mode. Stop the process with Ctrl+C to clear its in-memory report.

Open the local report ↗