Private · Agent runtime
Torok
Torok runs Canary Desk’s morning briefs and its AI analyst. You write the agent in ordinary Go, with your own functions as tools and your own types for the answer. It handles the tool loop and checks the answer against those types.
Torok saves progress as a run goes. A run that stops for a restart, or waits for a person to answer, picks up where it stopped without repeating what it already did. The code is private and the API is still changing.
Explainer film
How Torok works.
Where code sets the steps, where the model makes decisions, and how Torok saves the work between them.
- Checked rather than promised
- Race tests cover storage, sessions, typed repair, shared child limits, nested concurrency, approvals, ambiguous commits and event replay.
- Child-process tests kill the process after a real effect and before its receipt is saved, then require reconciliation without repeating the effect.
- A Lean 4 model of the acceptance boundary proves that a result is accepted only with an explained, passing check for every declared criterion. CI runs it and checks that removing the guard makes the proof fail.
- Model quality is a separate question. Scripted fixtures prove execution. What a live model does with the same tools needs real tasks and a person to grade them.
Incoming workAgent + toolsSaved progressHuman review