Agent trace exchange · pilot opening soon
Real agent work, licensed with consent.
Privacy-screened datasets of real AI coding-agent sessions, outcome-labelled where evidence exists.
For teams training and evaluating agents
Datasets of real agent work
- Real coding-agent sessions on real repositories, not synthetic tasks.
- Outcomes re-run in a sandbox where possible, with evidence.
- A consent record and rights dossier behind every trace.
For developers
Get paid for sessions you already ran
- You pick the projects. Nothing is selected by default.
- Secrets and personal data are redacted on your machine first.
- Paid when a dataset containing your trace is licensed.
What happens to a session
- 01
Export
Picked, redacted and consented on the contributor's machine.
- 02
Quarantine
Rescanned on the server with stricter detectors.
- 03
Checks
Authenticity, provenance, an offline re-run.
- 04
Review
A person decides every trace.
- 05
Dataset
Frozen versions with hashes and a rights dossier.
Waitlist
Join the pilot
We are starting with a small group on each side. Tell us which one you are on.