vLLM Metrics Monitor (vmm) is a lightweight dashboard that scrapes Prometheus metrics from vLLM, persists to SQLite, and serves a real-time web UI. Zero external dependencies — pure Python standard library.

🚀 Quick Start

# Install with uv
uv tool install vllm-metrics-monitor

# Or install with pip: `pip install vllm-metrics-monitor`

# Launch dashboard
vmm http://your-vllm:8000/metrics

Open http://localhost:8080 — that's it.

📖 Usage

vmm [URL] [OPTIONS]

Positional:
  URL                   vLLM Prometheus metrics endpoint
                        (default: http://localhost:8000/metrics)

Options:
  -p, --port PORT       Dashboard HTTP port (default: 8080)
  -i, --interval SEC    Scrape interval in seconds (default: 3)
  --retention HOURS     Data retention period (default: 720, i.e. 30 days)
  --db PATH             SQLite database path (default: ~/.vmm/data.db)
  --reset               Delete existing database and start fresh
  --debug               Enable debug logging

Examples

vmm http://vllm-server:8000/metrics -p 9090 -i 5
vmm http://vllm-server:8000/metrics --reset
vmm http://vllm-server:8000/metrics --retention 72 --db /data/vmm.db

Docker

# Build
docker build -t vmm .

# Run
docker run -d --network host vmm http://localhost:8000/metrics

# Or use docker compose
METRICS_URL=http://192.168.1.100:8000/metrics docker compose up -d

🏗️ Architecture

graph LR
    A[vLLM /metrics] -->|scrape every 3s| B[vmm]
    B --> C[Scraper Thread]
    C --> D[(SQLite<br/>~/.vmm/data.db)]
    B --> E[HTTP Server]
    E -->|JSON API| F[Browser<br/>Chart.js Dashboard]

📈 Monitored Metrics

License

MIT