Crate klieo_memory_sqlite 

SQLite-backed implementations of the three memory traits in klieo_core::memory.

MemorySqlite::open(path) is the capability-shaped default — opens a SQLite file (or ":memory:" for ephemeral), runs migrations for all three trait tables, defaults to DummyEmbedder, and returns three Arc<dyn …> handles ready to drop into klieo_core::AgentContext.

For a caller-supplied Embedder (e.g. the fastembed feature’s FastEmbedEmbedder), reach for MemorySqlite::new directly.

Quickstart

use klieo_memory_sqlite::MemorySqlite;

async fn example() {
    let mem = MemorySqlite::open(":memory:").await.unwrap();
    // mem.short_term, mem.long_term, mem.episodic are
    // Arc<dyn …> handles ready for AgentContext.
    let _ = mem;
}

Advanced wiring — custom embedder

use klieo_memory_sqlite::{MemorySqlite, DummyEmbedder};
use std::sync::Arc;

async fn example() {
    let mem = MemorySqlite::new(":memory:", Arc::new(DummyEmbedder)).await.unwrap();
    let _ = mem;
}

Features

• Default — three SQLite-backed memory traits + Embedder trait with DummyEmbedder (zero-vector). Linear-scan recall.
• fastembed — adds FastEmbedEmbedder (real CPU embeddings via ONNX runtime). Heavy first compile; model auto-downloads to OS cache dir on first call.
• sqlite-vec — adds a vec0 virtual-table index on long_term_facts_vec; LongTermMemory::recall uses k-NN MATCH instead of linear scan.
• test-utils — exposes FakeEmbedder for downstream test deps.

Limitations

• Sync SQLite under blocking pool. All trait methods spawn tokio::task::spawn_blocking so they don’t stall the async runtime; throughput is bounded by the blocking-pool size.

Re-exports

pub use short_term::SqliteShortTerm;

pub use episodic::SqliteEpisodic;

pub use long_term::SqliteLongTerm;

pub use factory::MemorySqlite;

Modules

embedder

Re-exports of the shared Embedder trait + dummy/fake impls.

episodic

SqliteEpisodic — EpisodicMemory over a SQLite table.

factory

MemorySqlite — umbrella factory wiring all three SQLite-backed memory traits to the same database file.

long_term

SqliteLongTerm — LongTermMemory over a SQLite table with linear-scan cosine recall (default) or sqlite-vec k-NN MATCH (under feature sqlite-vec).

short_term

SqliteShortTerm — ShortTermMemory over a SQLite table.

Structs

DummyEmbedder

Default embedder that returns zero vectors.

FakeEmbedder

Test-only embedder that hashes each input text into a deterministic vector. Identical texts produce identical embeddings, so cosine recall behaves predictably under test.

FastEmbedEmbedder

fastembed-rs–backed Embedder.

Constants

DEFAULT_DIM

Output dimension of DEFAULT_MODEL.

DEFAULT_MODEL

Default model: 384-dimensional AllMiniLML6V2.

Traits

Embedder

Compute embeddings for one or more texts.