Telemetry and performance intelligence for memory-driven AI systems. One install — full visibility into how your Mem0 stack performs.
Mem0 Analytics is the built-in monitoring and observability layer for the Mem0 framework. It measures, aggregates, and visualizes the complete lifecycle of memory interactions — from LLM calls to vector operations — with no manual setup or instrumentation required.
Once installed, it automatically:
- Captures latency, token usage, cache efficiency, and system metrics
- Aggregates performance data every 60 seconds
- Displays results through a real-time terminal dashboard
- Optionally syncs to PostHog for team dashboards
Install directly from PyPI:
pip install mem0-analyticsThen just use Mem0 as usual — analytics activates automatically.
To open the live dashboard, run:
mem0-dashboardMetrics are stored locally at:
~/.mem0_metrics.db
🛢️ To see the db, open the SQLite shell
sqlite3 ~/.mem0_metrics.db📊 List all tables
.tablesYou should see something like:
mem0_met mem0_kpiView the schema of a specific table
.schema mem0_kpior for the raw metrics:
.schema mem0_metPreview table contents
SELECT * FROM mem0_kpi LIMIT 10;See table info (columns and types)
PRAGMA table_info(mem0_kpi);| Category | Metrics | Description |
|---|---|---|
| Performance | avg_latency_ms, latency_p95, ttfr_ms |
Average, tail, and cold-start latency |
| Embeddings | avg_embed_latency |
Mean embedding generation time |
| Vector Store | avg_vector_latency, cache_effectiveness |
Query efficiency and cache utilization |
| System | cpu_percent, mem_used_mb |
Process-level system footprint |
| Reliability | success_rate, error_rate |
Operation health and stability |
A high-frequency dashboard rendered with rich, showing:
- Latency (mean and P95) by function
- Embedder and vector database performance
- Cache efficiency and request success rate
- Time-to-First-Response (TTFR)
- Live stability and health indicators
No external services. Runs fully offline.
For centralized analytics or multi-agent visibility:
export POSTHOG_API_KEY=<your_key>
export POSTHOG_HOST=https://app.posthog.comThe local aggregator automatically batches KPIs and publishes them to PostHog every 60 seconds.
Mem0 (LLM, Vector Store, Embedder)
│
├── analytics.py → auto-captures runtime metrics
│
├── ~/.mem0_metrics.db → local SQLite telemetry store
│
├── dashboard.py → live Rich terminal visualization
│
└── (optional) PostHog sync → team dashboards
Local-first by default — privacy-safe, transparent, and extensible.
Mem0 Analytics automatically works across all Mem0 integrations:
| Layer | Supported Backends |
|---|---|
| LLMs | OpenAI (gpt-5-nano, gpt-4o-mini), Ollama (smollm2), Claude, Gemini, LLaMA, DeepSeek, Groq, etc. |
| Vector Stores | Qdrant, ChromaDB, Weaviate, FAISS, Pinecone, Milvus, Redis, LanceDB |
| Embedders | OpenAI, Ollama, Hugging Face, Instructor, BGE, Sentence Transformers |
No adapters, no config. If it runs on Mem0 — it’s tracked.
Monitoring LLM pipelines shouldn’t need Grafana, SQL schemas, or complex telemetry setups. Mem0 Analytics makes performance transparent — giving developers instant insight into how memory operations behave, degrade, and optimize over time.
- ⚙️ Zero-config integration with Mem0
- 🪶 Lightweight — <1 ms overhead per call
- 📦 SQLite for local metrics
- 📊 Real-time Rich dashboard
- ☁️ PostHog cloud mode for teams
- 🔁 Rolling aggregation of KPIs every minute
- 🧩 Pluggable design — supports any backend
- Local SQLite metrics engine
- Live terminal dashboard
- PostHog sync
- Token & cost analysis
- Anomaly alerts (p95, cache, TTFR)
- Prometheus exporter
- Multi-agent benchmarking
Pull requests are open. Extend metric types, add providers, or improve visualizations.
Licensed under the MIT License.
See LICENSE for details.