Memory Architecture v4.0

LAR Labs Memory Architecture - Tiered context loading with SQLite persistence.

Installation

pip install -e .

Usage

Tiered Context Loading

from src.memory.tiered_loader import TieredLoader

loader = TieredLoader()

# Load at specific tier
context = loader.load_abstract(path)  # L0: brief summary
context = loader.load_overview(path)    # L1: overview
context = loader.load_full(path)        # L2: full content with chunking

# Auto fallback chain (L0 → L1 → L2)
context = loader.load_tiered(path)

Generate Abstracts

from src.memory.abstract_generator import AbstractGenerator

gen = AbstractGenerator()

# Generate from content
abstract = gen.generate_abstract(markdown_content)
overview = gen.generate_overview(markdown_content)

# Or from file
gen.save_summaries(Path("document.md"))

Persistence

from src.memory.persistence.fact_store import FactStore
from src.memory.persistence.audit_logger import AuditLogger

# CRUD operations
store = FactStore()
fact = store.create("key", "value", metadata={"source": "test"})
fact = store.get("key")
store.update("key", "new_value")
store.delete("key")

# Audit logging
logger = AuditLogger()
logger.log_insert("facts", fact.id, {"key": "key", "value": "value"})

Testing

pytest tests/unit/

Project Structure

src/memory/
├── __init__.py
├── config.py           # Configuration
├── tiered_loader.py    # L0/L1/L2 context loading
├── abstract_generator.py  # LLM summarization
└── persistence/
    ├── database.py     # SQLite connection
    ├── fact_store.py   # CRUD operations
    └── audit_logger.py # Audit trail