#local-first #n-count #ai-memory #community-knowledge

Memory Garden - Validated Community Memory

N-count validated knowledge for AI agents. Patterns that prove themselves through repeated use. Local-first, community-ready.

Lee Brown@leegitw

Install

openclaw skills install @leegitw/memory-garden

Memory Garden

Stop answering the same questions twice.

Memory Garden learns what works in your domain and brings validated knowledge to every conversation. Unlike simple memory tools, patterns here must prove themselves through repeated successful use (N-count convergence).

Community-validated. Local-first. You control what stays.

Why Memory Garden?

Problem	Solution
AI forgets what worked yesterday	Patterns persist and surface when relevant
No way to know if advice is reliable	N-count shows how many times a pattern proved useful
Memory silos per project	Federated curation lets you subscribe to trusted sources
Privacy concerns with cloud memory	Local-first: nothing leaves your machine unless you opt-in

Features

Feature	Default	Description
Search	ON	Augment queries with validated patterns
Extraction	OFF	Extract patterns from conversations (opt-in)
Validation	ON	Record when patterns help (increments N-count)
Sync	OFF	P2P pattern synchronization (opt-in)

How N-Count Works

text

N=1: Pattern recorded (observation)
N=2: Pattern confirmed (recurring)
N=3+: Pattern validated (reliable knowledge)

Higher N-count = more trustworthy. Patterns that keep helping rise to the top.

Quick Start

The skill automatically manages the Memory Garden daemon:

Starts daemon if not running
Uses port 18790 (or next available)
Health checks at /health

bash

# Verify installation
clawhub inspect memory-garden

# Check daemon status
curl http://127.0.0.1:18790/health

Configuration

Set via environment variables:

Variable	Default	Description
`MG_DAEMON_URL`	`http://127.0.0.1:18790`	Daemon URL (if external)
`MG_EXTRACTION_ENABLED`	`false`	Enable pattern extraction
`MG_EXTRACTION_CONFIRM`	`true`	Require human confirmation
`MG_SYNC_ENABLED`	`false`	Enable P2P sync
`MG_SEARCH_LIMIT`	`8`	Default search result limit

Platform Support

Platform	Status	Notes
macOS Apple Silicon	✅ Supported	arm64
macOS Intel	✅ Supported	amd64
Linux x86_64 (glibc)	✅ Supported	amd64, requires glibc 2.31+ (Ubuntu 20.04+, Debian 11+)
Linux ARM64	❌ Not supported	libSQL limitation - vote for ARM64 support
Windows	⚠️ WSL2 required	Native Windows not supported
Alpine/musl	❌ Not supported	Requires glibc (libSQL dependency)

Data Handling

Your patterns are your intellectual property. Local-first means your learning stays yours until you choose to share.

Patterns stored locally in ~/.memory-garden/
No data sent externally unless sync explicitly enabled
Extraction requires opt-in (MG_EXTRACTION_ENABLED=true)
All operations logged to ~/.memory-garden/logs/

Security

Daemon listens on localhost only (127.0.0.1)
No outbound network by default
Trust-gated tool access
HMAC-SHA256 webhook verification (for CI integration)
MIT License

Tools

Tool	Description	Default
`search_patterns`	Find validated patterns for your query	Enabled
`plant_pattern`	Record a new observation	Requires extraction
`validate_pattern`	Confirm a pattern helped (increments N-count)	Enabled
`get_category_effectiveness`	View fix success rates by category	Enabled

Integrations

Continuous AI Improver

Track which fix patterns actually get merged: