The Problem

File indexing isn't memory

OpenClaw's built-in memory indexes your markdown files. That's fine for recent context — but after 100 sessions, your daily logs are a haystack. You can't ask "what does this client prefer?" and get a precise answer. You get a wall of old conversation.

This toolkit adds fact-based memory. An LLM reads your session transcripts, extracts the knowledge that matters, classifies it, scrubs credentials, and stores structured facts in a vector database.

What It Adds

Beyond file indexing

	OpenClaw Built-in	+ This Toolkit
Session context	✓	—
Markdown file indexing	✓	—
Structured fact extraction	—	LLM extracts discrete facts from session noise
Long-term vector memory	—	Separate Qdrant store
Documentation knowledge base	—	Embed any docs, searchable
Client data isolation	—	Mandatory client_id on every memory
Credential scrubbing	—	API keys, tokens, emails redacted
GDPR-compliant deletion	—	Per-client erasure with audit log
Cross-agent bridge	—	Optional bridge to Multi-Agent Memory
Encrypted backups	—	GPG-encrypted Qdrant snapshots

Architecture

How it all connects

Hover over any node to learn what it does. Session data flows through consolidation into searchable memory that your agents can query.

Input

Session Logs

conversation_log.md

Docs Pipeline

Markdown → embeddings

Processing

Consolidation Engine

LLM extraction · JSON mode

cron batch dedup

Storage

Qdrant

Vector search

Encrypted Backups

GPG + rotation

Output

4 Agent Skills

store · query · delete · docs

↑ Hover a node to see details

Quick Start

Install and forget

git clone https://github.com/ZenSystemAI/openclaw-memory.git
cd openclaw-memory
cp .env.example .env
# Set OPENAI_API_KEY and QDRANT_API_KEY

# Install skills
cp -r skills/* ~/.openclaw/skills/

# Set up cron for automatic fact extraction
crontab -e
# 0 11,23 * * * bash ~/.openclaw/scripts/memory-consolidate.sh

Cross-Agent Bridge

Works with Multi-Agent Memory

If you run multiple agents across machines, the consolidation engine can automatically bridge cross-agent-relevant facts to Multi-Agent Memory's shared brain. Add two env vars and it just works.

View on GitHub See also: Multi-Agent Memory — the cross-machine shared brain that this toolkit can bridge to.