Core flows
Memory flow
Capture durable facts into signed memory modules, recall them by relevance, and curate what persists through proposal and review.
Memory is how a workspace remembers across sessions. Facts you capture become
modules in .seif/; the next agent reads them back at session start, so context
never dies when a conversation window closes.
contribute / gov memory add → fact written + MEMORY.md pointer
│
▼
memory query "<intent>" → most relevant facts recalled
│
memory propose → review → autonomous curation, owner-gated
│
gov memory audit → integrity: files ↔ MEMORY.mdCapture
# Append a signal to a contributions log
seif contribute decisions "Adopt OpenTimestamps for all release artifacts."
# Or add a curated, indexed memory entry
seif gov memory add --name adopt-ots --content "All release artifacts are timestamped with OpenTimestamps."Both write through to MEMORY.md, the capped index that AI tools load at session
start.
Recall
seif memory query "how do we timestamp releases" # relevance-ranked recall
seif memory surface # capped cross-AI bootstrapquery answers a specific intent; surface emits the bootstrap digest a cold
agent reads to orient.
Curate
seif memory propose ... # propose a memory for review
seif memory proposals # list what's awaiting review
seif memory review <id> # approve (absorb) or rejectAutonomous curation never writes durable memory unilaterally — proposals are owner-gated.
Keep it honest
seif gov memory audit # verify .seif/memory/ ↔ MEMORY.md integrityWhere to go next
- Quality & grounding — why grounded recall matters.
- Governance flow — decisions and deferrals that outlive a session.