Memorit

A semantic memory system for storing and retrieving conversational data using embeddings and concept extraction.

Features

• Hybrid Search: Combines vector similarity, conceptual matching, and keyword search
• Concept Extraction: Automatically extracts semantic concepts from conversations
• Clean Architecture: Well-defined separation between domain, storage, and AI layers
• Concurrent Processing: Async embedding and concept extraction with worker pools
• Pluggable Backends: Abstract interfaces for storage and AI services
• Fast Serialization: Uses mus-go for efficient binary serialization

Dependencies

• BadgerDB v4 - Embedded key-value store
• langchaingo - LLM and embedding integrations
• mus-go - Fast binary serialization
• ants - Worker pool implementation

Core Domain (core/)

Pure domain models with zero external dependencies:

• Domain entities: ChatRecord, Concept, ConceptRef, ID, SpeakerType
• Business validation rules
• Domain-specific errors

Storage Layer (storage/)

Repository pattern with BadgerDB implementation:

• ChatRepository: Chat record operations
• ConceptRepository: Concept operations
• VectorSearcher: Vector similarity search
• Thread-safe with context support

AI Layer (ai/)

Provider abstraction for AI services:

• Embedder: Generate vector embeddings from text
• ConceptExtractor: Extract semantic concepts
• AIProvider: Unified interface for AI operations
• Includes OpenAI and mock implementations

Ingestion Pipeline (ingestion/)

Concurrent processing pipeline:

• Stores chat records
• Generates embeddings asynchronously
• Extracts and assigns concepts in parallel
• Worker pools for maximum throughput

Search Engine (search/)

Multi-stage hybrid search:

• Semantic search via vector embeddings
• Conceptual search using extracted concepts
• Keyword matching with stop-word filtering
• Ranked results with relevance scoring

Quick Start

Prerequisites

• Go 1.25.3+
• Task for build automation

Building

# Build all binaries
task build

# Run tests
task test

# Generate serialization code
task generate

# Run static analysis
task vet

Usage

Seed the database:

# Use built-in test data
./bin/seeder

# Load from file
./bin/seeder -src data.txt

Search conversations:

./bin/searcher

Development

Running Tests

# All tests
task test

# Specific package with coverage
go test -v -cover ./core/
go test -v -cover ./storage/badger/

Code Generation

Memorit uses mus-go for binary serialization. Delete core/records_musg.gen.go then use the generate task to regenerate serializable models.

License

Licensed under the Apache License 2.0. See LICENSE for the full license text and NOTICE for third-party attributions.