dvcrn LanceDB Memory (Windows-ready)
Use this skill for long-term memory storage and retrieval in OpenClaw/Clawdbot.
What this skill provides
- Cross-platform path handling (Windows/macOS/Linux)
- Persistent memory table in LanceDB
- Text retrieval API compatible with existing memory hooks
- Async provider (
search/add/get_recent) for OpenClaw plugins
Install dependencies
Run in your project environment:
pip install lancedb pandas pyarrow
Storage path
Default path logic:
- Windows:
D:\clawtest\memory\lancedb - macOS/Linux:
~/.clawdbot/memory/lancedb
Override with:
OPENCLAW_LANCEDB_PATH(highest priority)CLAWTEST_ROOT(Windows base directory)
Python API
from final_memory import add_memory, search_memories, get_all_memories
mid = add_memory("Use GLM embeddings for semantic recall", {"type": "decision"})
hits = search_memories("GLM", limit=5)
all_items = get_all_memories()
Async provider API (for OpenClaw integration)
from clawdbot_memory import memory_provider
results = await memory_provider.search("OpenClaw", limit=10)
memory_id = await memory_provider.add("Important preference", {"importance": 0.9})
recent = await memory_provider.get_recent(limit=20)
Notes
- This implementation keeps behavior stable with the original skill: add, query, list.
- Retrieval is text-based by default (robust and dependency-light).