Expand description
The memory domain model — the vocabulary shared by every stage of the pipeline (capture → staging → retrieval → reconciliation → promotion).
Types here are deliberately serialization-stable: they are the schema of the OKF front matter, the event log, and the structured-output contracts handed to out-of-band extraction models.
Structs§
- Canonical
Memory - A reconciled, durable memory — the canonical unit of the OKF repository.
- Canonical
Predicate - A canonicalized predicate name such as
dietary_identity. - Entity
Ref - A reference to the subject of a memory.
- Evidence
Counters - How much evidence stands behind a memory.
- Fact
Fingerprint - A deduplication key for “the same fact stated twice”.
- Memory
Observation - A structured interpretation of one specific user statement.
- Memory
Source - Where a memory came from.
- Privacy
Metadata - User-facing data rights for a record.
- Retrieval
Metadata - The fields a memory is matched on during lexical retrieval.
- Temporal
Metadata - Validity and freshness metadata.
- Transcript
Evidence - The evidence span an observation was drawn from.
Enums§
- Explicitness
- How directly the user stated the thing being remembered.
- Memory
Kind - What sort of thing a memory records.
- Memory
Status - Lifecycle state of a canonical memory record.
- Memory
Value - The value side of a memory triple.
- Mutation
Intent - An explicit user instruction about memory itself.
- Proposed
Persistence - What the extractor proposes should happen to a candidate over time.
- Sensitivity
Class - Privacy classification, gating automatic promotion.
- Speaker
Attribution - Who actually said the thing an observation was drawn from.
- Temporal
Scope - How long a fact is expected to hold.
Functions§
- normalize_
token - Lowercase a string and collapse everything non-alphanumeric to single spaces.
- stable_
hash - A stable, dependency-free 64-bit content hash rendered as hex.