Core: public API map

August 17, 2026 · View on GitHub

Where to find what in velesdb_core. This is an import map, not a signature reference — full signatures, argument names and return types are generated on docs.rs/velesdb-core and are the authority.

Moved out of crates/velesdb-core/README.md to keep that file under the 400-line documentation budget.

Database, collections, points

use velesdb_core::{
    Database,            // database handle (open, create/get collections, execute_query)
    VectorCollection,    // vector collection (typed handle)
    GraphCollection,     // graph collection (typed handle)
    MetadataCollection,  // metadata-only collection (typed handle)
    AnyCollection,       // type-erased handle from Database::get_any_collection
    Point,               // id + vector + optional JSON payload
    SearchResult,        // point + score
    ScoredResult,        // id + score (payload-free fast path)
    DistanceMetric,      // Cosine, Euclidean, DotProduct, Hamming, Jaccard
    StorageMode,         // Full, SQ8, Binary, ProductQuantization, RaBitQ
    Error, Result,       // error types
};

Collection itself is pub(crate) since v1.13 — the typed split above is the public API.

Graph engine

use velesdb_core::{
    GraphEdge, GraphNode, EdgeType, NodeType, GraphSchema,
    TraversalConfig, TraversalPath, TraversalResult,
};

VelesQL

use velesdb_core::velesql::Parser;   // parse a VelesQL statement
// then run it with Database::execute_query

Sparse vectors and fusion

use velesdb_core::sparse_index::SparseVector; // (index, weight) pairs
use velesdb_core::FusionStrategy;             // RRF, RelativeScore, ...

Streaming ingestion

use velesdb_core::collection::streaming::{
    StreamIngester,     // bounded-channel ingestion pipeline
    StreamingConfig,    // buffer size, batch size, flush interval
    BackpressureError,  // BufferFull, NotConfigured, DrainTaskDead
};

Agent memory (requires the `persistence` feature)

use velesdb_core::agent::{
    AgentMemory,        // unified interface (semantic + episodic + procedural)
    SemanticMemory,     // long-term knowledge
    EpisodicMemory,     // event timeline with temporal queries
    ProceduralMemory,   // learned patterns with reinforcement
    ProcedureMatch,     // recall result with confidence and steps
    EvictionConfig,     // TTL and eviction policy
    ExpireResult,       // auto_expire() report
    SnapshotManager,    // versioned snapshot persistence
    TemporalIndex,      // B-tree temporal index, O(log N) time queries
};

Indexes

use velesdb_core::{
    HnswIndex,       // HNSW index
    HnswParams,      // index parameters
    SearchQuality,   // Fast, Balanced, Accurate, Perfect, Custom, Adaptive
    VectorIndex,     // index trait
};

Query plan cache

use velesdb_core::cache::{
    CompiledPlanCache,  // two-tier LRU cache for compiled plans
    PlanCacheMetrics,   // hit/miss counters with hit_rate()
    PlanKey,            // deterministic key (query hash + write generation)
};

Filtering

use velesdb_core::{Filter, Condition};

Quantization

use velesdb_core::{QuantizedVector, BinaryQuantizedVector, QuantizationConfig};

Retrieval-quality metrics

use velesdb_core::{recall_at_k, precision_at_k, mrr, ndcg_at_k};