A read-only web dashboard for observing mem0 memory stores.
Self-hosted mem0 has no built-in way to see what your AI agents actually remember. Memories live inside a Qdrant vector database as opaque embeddings and payloads — there's no UI to browse, search, or monitor them.
This dashboard gives you visibility into that black box: what each agent knows, how memories grow over time, and a way to query them using natural language.
- Memory Browser — paginated list of all stored memories, filterable by agent, with client-side search
- Agent Tabs — per-agent memory counts and quick switching (auto-detected from Qdrant data)
- Activity Feed — recent memory activity across all agents
- Query Explorer — natural language search via OpenAI embeddings + Qdrant vector similarity
- Performance — Qdrant telemetry, search/write latency, time-series activity charts
- Settings — read-only view of mem0 + Qdrant + dashboard configuration
- Auto-refresh — all data refreshes on a configurable interval
- Dark mode only — designed for terminal-adjacent workflows
cp .env.example .env
# Edit .env with your Qdrant URL and (optionally) OpenAI key
docker compose up -d --build
# Dashboard available at http://localhost:8765npm install
cp .env.example .env
# Edit .env
npm run dev
# http://localhost:3000All configuration is via environment variables (see .env.example):
| Variable | Required | Default | Description |
|---|---|---|---|
QDRANT_URL |
Yes | http://localhost:6333 |
Qdrant REST API endpoint |
QDRANT_COLLECTION |
Yes | openclaw-memories |
Collection name |
OPENAI_API_KEY |
For Query Explorer | — | Must use same embedding model as your mem0 stack |
OPENAI_EMBEDDING_MODEL |
For Query Explorer | text-embedding-3-small |
Embedding model |
DASHBOARD_PORT |
No | 8765 |
External port (Docker) |
AGENTS |
No | auto-detect | Comma-separated agent list |
REFRESH_INTERVAL |
No | 60 |
Auto-refresh interval in seconds |
- Next.js 14 (App Router) — single container, standalone output
- TypeScript
- shadcn/ui + Tailwind CSS + Framer Motion
- @qdrant/js-client-rest — direct Qdrant access
- openai — embeddings for Query Explorer
- Docker — Node.js 20 Alpine, multi-stage build
The Qdrant collection may contain records from two schema generations (pre/post mem0 migration). The dashboard handles both transparently:
Old: user_id, created_at (no runId)
New: userId, createdAt, runId
All queries use should (OR) filters to match both field names.
- Read-only — the dashboard never writes, modifies, or deletes memories
- No auth — designed for LAN-only deployment
- Sorting in Node.js — Qdrant can't sort on string timestamp fields
MIT