Deep dives from building an event-first, multi-user AI platform.
Search covers titles, summaries, tags and full article text.
An app owns a realm — task:, mem:, cnv: — with its own schema, search, actions, and rendering. Named services let any hosted agent enter that realm without learning any of its private domain rules. The agent gets one generic interface; the provider remains the owner of meaning. This Deep piece walks the four agent surfaces, the pull/read materialization p...
An app can stay an ordinary service — or expose its domain as a realm that other apps, composed surfaces, and configured agents use without copying its database. Each surface adds reach without a rewrite.
Objects from many apps sit together on one board without merging their databases. Every durable card is a proxy: canvas owns the card; the provider named by the ref owns the object.
One ordered block of input, assembled fresh for every call: a per-agent instruction envelope, tool and skill catalogs, a growing timeline, and a volatile tail — each part with a provenance.
The attention zone at the tail of the model input: the sources pool and the live view — never cached, rebuilt every round, the one place checked first.
How the model input is cached and kept fresh: four moving cache points, the never-cached tail, marked cold turns, TTL pruning, and compaction.