Engram — Graph Memory for OpenClaw Agents

Engram is a temporal knowledge graph memory system for OpenClaw agents. It extracts entities, facts, relationships, and emotions from session logs, stores them in a Kuzu graph database, and injects relevant context into every conversation turn.

Features

• Multi-agent memory isolation — Each agent's facts scoped by agent_id
• Parallel ingest — 6+ workers for LLM extraction (~10 files/min)
• Context engine plugin — Injects graph facts into OpenClaw agent context per turn
• Entity deduplication — Normalize and merge duplicate entities
• Interactive dashboard — Sigma.js graph visualization with per-agent filtering
• Hourly sync — Cron-based export → ingest → briefing pipeline
• Semantic + graph search — Kuzu graph queries + Chroma vector embeddings

Architecture

Session JSONL → Export → Markdown → LLM Extraction → Kuzu Graph DB
                                                          ↓
                                          Context Engine Plugin → Agent turns
                                                          ↓
                                              Dashboard (optional)

Components

Directory	Description
*.py (root)	Core: ingest, query, schema, export, dedup, consolidation
dashboard/	FastAPI + Sigma.js visualization with agent filtering
extensions/context-engine/	OpenClaw plugin for context injection
skills/engram/	Setup guide as an OpenClaw skill

Quick Start

See skills/engram/SKILL.md for full setup instructions.

# Install dependencies
python3 -m venv .venv-memory
source .venv-memory/bin/activate
pip install kuzu chromadb

# Export sessions and run ingest
python export_sessions.py
python ingest.py --workers 6

# Query
python context_query.py query "search terms" --agent main

Schema

Nodes: Entity, Fact, Episode, Emotion, SessionState
Relationships: RELATES_TO, CAUSED, PART_OF, MENTIONED_IN, EPISODE_EVOKES, ENTITY_EVOKES, DERIVED_FROM, ABOUT

Every node has an agent_id field for multi-agent isolation.

Dashboard

cd dashboard
npm install && npm run bundle
pm2 start ecosystem.config.js
# → http://localhost:3847

Features: graph visualization, per-agent filtering, entity search, node detail view, connection explorer.

Requirements

• Python 3.10+
• Node.js 18+ (dashboard only)
• Kuzu (via pip)
• LLM API access (xAI/Grok by default, configurable)
• OpenClaw (for context engine plugin)

License

MIT