| name | mindgardener |
| description | Local-first long-term memory for autonomous agents. Built for OpenClaw. Creates wiki knowledge graph from conversations, scores events by surprise, detects conflicts, and assembles token-budget context. Complements OpenClaw's built-in memory_search. |
| metadata | {"clawdbot":{"requires":{"bins":"[Truncated]"},"install":["[Truncated]"]}} |
MindGardener ๐ฑ
Your agents forget everything. This fixes it.
Built for OpenClaw. Complements the built-in memory_search tool.
How It Complements OpenClaw
| OpenClaw built-in | MindGardener adds |
|---|
| Search existing memory | Create memory from conversations |
| Manual MEMORY.md edits | Auto-extract entities โ wiki pages |
| Flat text search | Knowledge graph (triplets + wikilinks) |
| โ | Surprise scoring (unexpected = important) |
| โ | Conflict detection (new info vs old) |
| โ | Multi-agent sync |
Features (v1.1)
- ๐ Provenance tracking โ know where every fact came from
- โ๏ธ Conflict detection โ flags when new info contradicts old
- ๐ Auto-injection โ context ready at session start
- โฐ Temporal decay โ old facts fade unless reinforced
- ๐ Concurrency โ file locks for multi-agent safety
- ๐ฎ Associative recall โ follow wikilinks + graph traversal
- ๐ Confidence levels โ not all facts are equally reliable
- ๐ค Multi-agent sync โ merge per-agent memories to shared
Quick Start
pip install mindgardener
garden init
Add to your nightly cron:
garden extract && garden surprise && garden consolidate
Add to session start (BOOTSTRAP.md or heartbeat):
garden inject --output RECALL-CONTEXT.md
What Changes From Default OpenClaw?
- New folder:
memory/entities/ (wiki pages)
- New file:
graph.jsonl (knowledge triplets)
- New file:
RECALL-CONTEXT.md (auto-generated context)
- New file:
garden.yaml (configuration)
Everything is markdown files. No database. Works offline.
Requirements
- Python 3.10+
- No external APIs required
- For fully local: use
garden init --provider ollama
Links
