Brain & memory
A Brain that remembers what matters, and cites where it learned it.
Most agent memory is a pile of text that gets searched. The amoOS Brain is compiled: a living wiki of projects, decisions, concepts and people that every agent reads before it works.
Compiled, not piled
Typical agent memory
- Every note stays a separate chunk, forever.
- Old and new claims are blended silently.
- Agents re-read the same raw text every session.
- No way to tell where an answer came from.
The amoOS Brain
- New evidence rewrites one canonical page per topic.
- Conflicts are flagged for review, never blended.
- Agents get compiled context first, raw evidence on demand.
- Every page keeps links to the sources that shaped it.
How it works
01
Capture
Sessions, handoffs, documents, articles, videos and posts.
02
Compile
Sources become summaries and rewrite the canonical pages they touch.
03
Review
Contradictions, stale claims and orphans are flagged for a human.
04
Recall
Compiled context first, raw sources when exact evidence matters.
05
Reuse
Relevant lessons ride along with the next dispatch.
Every page knows where it came from, and what’s wrong with it.
Pages carry versions, provenance and open review flags. Orphans, contradictions and stale claims surface in a review queue instead of quietly degrading recall.
Operating amoOS daily
Telegram is the remote control, the laptop is the cockpit, the always-on machines are the engine room, and the Brain is the ledger everything writes back to.
Linked pages
5
Sources
2
Open flags
1
Provenance
Review flag · orphan. No inbound links yet, so this page is hard to reach from the graph.
Why it matters
No rediscovery
Cross-project transfer
Cited answers
Measured recall
Built on PostgreSQL with pgvector: structured pages and revisions, with hybrid retrieval over sources.