Loop
Tell Agent OS what 'done' means — it builds, independently checks the result, fixes what failed, and repeats until it passes or needs you.
Tell Agent OS what “done” means. It builds, independently checks the result, fixes what failed, and repeats until it passes or needs you.
A builder does the work; a separate judge grades it. The builder never grades its own work — if the judge fails the result, the notes go back and it tries again. Not to be confused with Workflows: a workflow defines how a job is done; Loop keeps executing and reviewing a job untilit’s actually done.
- 1 Define done
- 2 Build
- 3 Judge
- 4 Improve
- repeats until it passes, hits max rounds, or needs you
Running many rounds uses a lot of tokens, so the builder and the judge both default to free models, and the judge falls back to your local model if it’s ever throttled. Connect Nous Portal — run hermes portal — for more free models in both menus. Results are saved to your vault under Agentic OS/Loops.
No loops yet
Define what “done” means above and run a loop. Passing builds land here as a pass/fail history you can reopen.