Expand description
Vector indexes for caller-owned semantic retrieval.
This capability is independent from MemoryStore: callers own document
admission, embedding generation, lifecycle, and result fusion.
Caller-owned vector indexing primitives.
The module deliberately does not generate embeddings or infer semantic
equivalence. A caller supplies admitted vectors and owns their lifecycle.
The in-memory backend is dependency-free; the sqlite feature adds local
durability and cross-process revision compare-and-swap.
Structs§
- InMemory
Vector Index - Exact, session-ephemeral vector index backed by immutable partition blocks.
- Vector
Index Change Token - Exact revision evidence scoped to one vector-index history.
- Vector
Index Descriptor - Immutable vector shape and logical resource budgets.
- Vector
Index Observation - One self-consistent observation of a vector index revision.
- Vector
Index Status - Current logical size and published revision of an index.
- Vector
Record - One caller-owned vector record.
- Vector
Revision - Monotonic revision of one vector index.
- Vector
Search Hit - One ranked vector match.
- Vector
Search Request - A bounded exact-vector query.
- Vector
Search Result - Ranked matches and the exact immutable revision that produced them.
Enums§
- Vector
Budget Resource - Resource whose configured budget rejected a mutation.
- Vector
Index Error - Typed failures from vector admission, mutation, and search.
- Vector
Metric - Similarity function used by an index.
- Vector
Mutation Consistency - Strongest mutation ordering contract exposed by one vector backend.
- Vector
Normalization - Vector transformation applied at admission and query time.
Constants§
- VECTOR_
INDEX_ CHANGE_ TOKEN_ PROFILE_ V1 - Stable profile for exact vector-index history tokens.
Traits§
- Vector
Index - A bounded vector index whose content and lifecycle are owned by its caller.