Memory Tree — Three-Scope Structured Memory

A structured memory system with three scopes (source, topic, global) that automatically organizes, summarizes, and indexes agent memory. Never lose context again.

Architecture

Source (raw) → Topic (summarized) → Global (cross-topic knowledge)
   134 files         13 files              6 files

Quick Start

from memory_tools import MemoryTools

tools = MemoryTools(workspace="/path/to/workspace")

# Store a memory
tools.store("User prefers dark mode for all UI", topic_hint="preferences")

# Recall memories
results = tools.recall("dark mode", scope="global")

# Seal source → topic summaries
tools.seal()

# Check status
status = tools.status()
# → {"source_files": 134, "topic_files": 13, "global_files": 6}

Features

• Three-scope hierarchy: Source → Topic → Global
• Automatic sealing: Raw memories compressed into topic summaries
• Topic extraction: Automatic categorization by content
• Index management: Full-text search across all scopes
• Budget enforcement: Topic files \x3C 2000 tokens, Global \x3C 5000 tokens
• Idempotent sealing: Safe to run multiple times

Testing

python -m pytest test_memory_tree.py -v
# 15 tests passing

License

MIT