Chat for the first move. Flows and Offices for the next. SpiderMesh when bounded work needs to travel, return, and be reviewed.
Verityasksyoutotrustamechanismratherthanapromise.Atenantfilterthatfailsclosed.Adigestthateithermatchesordoesnot.Aplanlimittheserversettlesagainstaverifiedidentity.
Choose a bounded task, route it across available workers, then review the output. Verity keeps the next action visible instead of treating a model reply as the final word.
Each memory is tagged from five lanes (entity, intent, domain, temporal, lexical) and every tag carries a computed importance.
Recall matches tags first, drops weak-only overlap, then scores only what survived. A per-tenant registry aliases near-duplicates and bridges siblings, so asking about AI reaches the memory filed under large language model.
A conductor plans, workers run in parallel, an inspector converges the result.
Offices are standing teams. They checkpoint their state to a file, fire on a cron lease, and resume where they stopped.
Both run on the four Verity model tiers or on whichever providers you have configured. The models stay yours.
Verity is source-available and runs on your own hardware. Start in the browser, or clone it and point it at your models.