Crate aimem_core 

aimem-core

Give your AI a memory — reusable core library.

aimem-core is a pure-Rust library that stores, retrieves, and searches text memories using Turso (an in-process SQLite engine with native vector support) as the sole storage backend.

Feature overview

Feature	Implementation
Vector storage	Turso vector32() BLOB column
Semantic search	vector_distance_cos() — SimSIMD-accelerated cosine
Full-text search	SQL LIKE scan (Turso FTS index optional)
Embedding generation	Gemini 2.0 (Remote) or fastembed (Local)
Knowledge graph	Temporal entity-relationship triples in SQLite
Graph traversal	BFS over wing/room adjacency — no graph DB needed

Architecture: 4-Layer Memory Stack

┌─────────────────────────────────────────────────────────┐
│  L0  Identity text           ~100 tokens   always loaded │
│  L1  Essential story         ~500-800 tok  always loaded │
│  L2  On-demand recall        ~200-500 tok  loaded on ask │
│  L3  Deep semantic search    unlimited     full memory   │
└─────────────────────────────────────────────────────────┘

Re-exports

pub use config::Config;

pub use convo::ConvoMineStats;

pub use convo::ConvoMiner;

pub use db::AimemDb;

pub use db::EmbeddingStoreProfile;

pub use embedder::Embedder;

pub use embedder::Gemini2Embedder;

pub use embedder::LocalEmbedder;

pub use graph::AimemGraph;

pub use graph::TraversalNode;

pub use knowledge::KnowledgeGraph;

pub use layers::MemoryStack;

pub use miner::Miner;

pub use search::Searcher;

pub use types::ContentPart;

pub use types::Drawer;

pub use types::DrawerMeta;

pub use types::Entity;

pub use types::RoomNode;

pub use types::SearchResult;

pub use types::Triple;

pub use types::Tunnel;

Modules

config

Configuration — mirrors Python’s config.py

convo

Conversation miner — mine chat exports into AiMem.

db

AimemDb — Turso connection wrapper + schema bootstrap.

embedder

Embedder — text-to-vector abstraction and implementations.

extractor

Memory extractor — heuristic keyword extraction, no LLM required.

graph

AiMem graph — traverse rooms and wings in the memory store.

knowledge

Knowledge graph — temporal entity-relationship graph.

layers

4-layer memory stack — mirrors Python’s layers.py.

miner

File miner — scan a project directory and file chunks into AiMem.

prelude

Curated high-level imports for typical AiMem integrations.

search

Semantic search and FTS over the AiMem store.

types

Shared data types.