engram

Unified multi-agent memory for OpenClaw.

An elephant never forgets. Neither should your agents.

Engram is a shared memory layer that any OpenClaw agent can plug into. Cloud-synced through Convex with native vector search and full agent-native architecture.

What It Does

• Store atomic facts, entities, and conversations across agents
• Recall semantically — vector search finds what matters, not just what matches
• Share memory between agents with scoped access control
• Decay gracefully — old memories fade but never disappear
• Sync across devices — Mac Mini, MacBook Air, MacBook Pro all see the same brain
• Enrich automatically — embeddings, importance scoring, lifecycle management
• Recall three ways — semantic vector search + symbolic text + graph expansion, fused via Reciprocal Rank Fusion

Quick Start

OpenClaw (one-liner)

This does everything: deploys Convex → builds plugin → patches OpenClaw config → bootstraps existing memory files → verifies recall. Restart OpenClaw after.

Auto-Memory Discovery

Engram auto-ingests existing memory files from your workspace. The installer bootstraps all .md files from ~/.openclaw/workspace/memory/ on first run, then an hourly cron syncs new/modified files.

State tracked in memory/.engram-sync-state.json — only changed files re-ingested.

Add Engram to your agent (easy guide)

Use Engram from Claude Code, OpenClaw, or OpenCode in a few steps.