Zurück

How Astell finds and surfaces loops

The ingestion model, the storage layers, permissions, and how Astell shows up inside Claude or ChatGPT.

Ingest, not retrieve

Most AI tools retrieve: they fetch your data live every time you ask, with calls to Slack, Gmail, Notion. Slow, lossy, no memory of what came before. Astell ingests: it pulls your data in once and keeps it in sync. Every artifact lives in the relational, vector, and graph stores at the same time, so queries come back fast with the shape already there.

Three storage layers: relational, vector, graph

Every artifact lives in three places at once:

1. Relational store, structure. This is a ticket, that's an email. Where the type lives.
2. Vector store, meaning. "Maya's check-in" and "the audit follow-up" are about the same thing. Where similarity lives.
3. Graph, links and time. Which thread led to which ticket led to which decision. Where the shape of a loop lives.

Search tools live in vectors. Databases live in relations. Knowledge graphs live in graphs. Holding all three at once is what lets Astell carry the full shape of a loop and answer fast. The shape is computed before you ask, not after.

Living memory: the knowledge graph

Astell maintains a knowledge graph of everything connected to everything else, which thread led to which ticket led to which decision, who wrote what when, what changed and when it changed. The Mar 14 Slack thread that became MAYA-204 and the Jul 1 email about it are linked nodes, not separate searches. Every change is remembered. Most systems overwrite; Astell doesn't.

Permission mirroring

Astell mirrors source-system permissions identity by identity: they're captured as content syncs and re-checked on dedicated cycles, so new grants and revocations follow quickly. What you can see in Slack, in Gmail, in Notion, you see here. What you can't see, you don't. Tenant isolation is the outer boundary; per-user permission mirroring is the inner one. Permissions aren't bolted on. They propagate as data is ingested.

Unified search

Unified search returns relevant results across every tool you've connected: Slack, Gmail, Linear, Notion, Drive, GitHub, Calendar, ranked by relevance. Results link directly to the source artifact, so you can verify the snippet in its original context. The search index is the same one the Loops dashboard reads from.