Crate mnemefusion_core 

MnemeFusion Core

A unified memory engine for AI applications—“SQLite for AI memory.”

MnemeFusion provides four-dimensional memory indexing (semantic, temporal, causal, entity) in a single embedded database file with zero external dependencies.

Quick Start

use mnemefusion_core::{MemoryEngine, Config};

// Open or create a database
let engine = MemoryEngine::open("./brain.mfdb", Config::default()).unwrap();

// Add a memory
let embedding = vec![0.1; 384]; // Your embedding vector
let id = engine.add(
    "Project deadline moved to March 15th".to_string(),
    embedding,
    None,
    None,
    None,
    None,
).unwrap();

// Retrieve a memory
let memory = engine.get(&id).unwrap();

// Close the database
engine.close().unwrap();

Architecture

MnemeFusion is built on:

• redb: ACID-compliant embedded storage
• usearch: High-performance vector similarity search (HNSW)
• petgraph: Graph algorithms for causal and entity relationships

All data is stored in a single .mfdb file, making it easy to backup, version, and deploy.

Re-exports

pub use config::Config;

pub use error::Error;

pub use error::Result;

pub use graph::CausalEdge;

pub use graph::CausalPath;

pub use graph::CausalTraversalResult;

pub use graph::EntityQueryResult;

pub use graph::GraphManager;

pub use index::TemporalIndex;

pub use index::TemporalResult;

pub use index::VectorIndex;

pub use index::VectorIndexConfig;

pub use index::VectorResult;

pub use ingest::EntityExtractor;

pub use ingest::SimpleEntityExtractor;

pub use memory::contextualize_for_embedding;

pub use memory::first_person_to_third;

pub use memory::EmbeddingFn;

pub use memory::MemoryEngine;

pub use memory::ScopedMemory;

pub use query::AdaptiveWeightConfig;

pub use query::FusedResult;

pub use query::FusionEngine;

pub use query::IntentClassification;

pub use query::IntentClassifier;

pub use query::IntentWeights;

pub use query::QueryIntent;

pub use query::QueryPlanner;

pub use slm::SlmConfig;

pub use trace::Trace;

pub use trace::TraceRecorder;

pub use trace::TraceStep;

pub use trace::TraceValue;

pub use types::AddResult;

pub use types::BatchError;

pub use types::BatchResult;

pub use types::Entity;

pub use types::EntityId;

pub use types::FilterOp;

pub use types::Memory;

pub use types::MemoryId;

pub use types::MemoryInput;

pub use types::MetadataFilter;

pub use types::Source;

pub use types::SourceType;

pub use types::Timestamp;

pub use types::UpsertResult;

pub use types::NAMESPACE_METADATA_KEY;

pub use types::SOURCE_METADATA_KEY;

Modules

config

Configuration types for MnemeFusion

embedding

Embedding model integration for automatic text vectorization.

error

Error types for MnemeFusion

graph

Graph module for causal and entity relationships.

index

Index layer - vector, temporal, and keyword indexing

ingest

Ingestion module for entity extraction and memory preprocessing

memory

Memory engine - main API entry point

query

Query module for intent classification and multi-dimensional retrieval

slm

Small Language Model (SLM) integration for semantic intent classification

storage

Storage layer for MnemeFusion

trace

Pipeline observability / tracing

types

Core types for MnemeFusion

util

Utility functions

Macros

trace_begin

Begin a named step on an Option<&mut TraceRecorder>. No-op when the recorder is None.

trace_record

Record a key-value pair on an Option<&mut TraceRecorder>. No-op when the recorder is None.