sandclaw-memory v0.1.0 "Remembering AI"

The first public release of sandclaw-memory — a zero-dependency, self-growing tag-dictionary RAG library for Python.

Install

pip install sandclaw-memory

Highlights

• 3-Layer Temporal RAG: L1 (3-day session logs) → L2 (30-day AI summaries) → L3 (permanent SQLite + FTS5 archive)
• Self-Growing Tag Dictionary: Stage 1 keyword_map (instant, free) + Stage 2 AI queue (async) — costs decrease over time
• Intent-Based Depth: CASUAL → L1 only, STANDARD → L1+L2, DEEP → L1+L2+L3
• 5 AI Callbacks: tag_extractor (required), promote_checker, depth_detector, duplicate_checker, conflict_resolver
• Zero Dependencies: Python stdlib + sqlite3 only
• Cross-Platform: Windows, macOS, Linux — Python 3.9-3.13

Quick Start

from sandclaw_memory import BrainMemory
with BrainMemory(tag_extractor=my_ai_func) as brain:
 brain.start_polling()
 brain.save("User loves Python and React", source="archive")
 print(brain.recall("what does the user like?"))

Links

• PyPI: https://pypi.org/project/sandclaw-memory/0.1.0/
• Docs: Architecture | API Reference
• Examples: basic_usage.py | with_openai.py | with_anthropic.py

Full Changelog

https://github.com/kokogo100/sandclaw-memory/commits/v0.1.0