Recorded evidence

Public interactive demoActions replay reviewed evidence summaries and never consume project credits.

Private workspace

Welcome back

Sign in with this instance's application token. Your Nebius API key never belongs here.

Working product

Investigate once.
Protect the next change.

Run the reviewed Werkzeug scenario, inspect execution evidence, create a Memory Capsule, then recall it for a future refactor.

Start here

Bring a repository and describe the behavior worth protecting.

Investigate a Python project

Public GitHub repositoryGit ref

or

Upload a repository ZIP up to 10 MBWhat should remain true?

Public imports use GitHub's fixed archive host. ZIP is available for private or offline source.

Repository memory

Use semantic recall · one budgeted embedding batch

Background investigations

Recent investigations

Recorded live evidence

Completed real runs. Viewing them spends no credits.

Reviewed demonstration scenarios

Repair modeScenario

Python + pytest

Prewritten pricing fixture · two attempts maximum · independent boundary checks

This demonstration uses prewritten patches and tests.

↻

A living record of intent

Investigate a failure once. Preserve what fixed it.
Replay that knowledge when future agents touch related code.

01 Learn02 Remember03 Recall04 Replay

Evidence covers the selected checks, not universal correctness.Continue to MCP and CI →