The open-source observability platform for LLM applications. Trace every call. Evaluate quality. Detect vulnerabilities.
╔╦╗┌─┐┌┬┐┌─┐┬ ┬┬ ┬┌┐ ╔═╗╦ ╦
║║║├┤ │ ├─┤├─┤│ │├┴┐ ║ ║ ║
╩ ╩└─┘ ┴ ┴ ┴┴ ┴└─┘└─┘ ╚═╝╩═╝╩
AI Agent Observability
╭──────────────────── 📊 Evaluation Results ────────────────────╮
│ Evaluator Score Bar Rating │
│ completeness 0.92 ████████████░░ Excellent │
│ latency 0.78 ██████████░░░░ Good │
│ cost 0.95 █████████████░ Excellent │
│ tool_usage 0.88 ███████████░░░ Good │
╰───────────────────────────────────────────────────────────────╯
✅ Overall Score: 88.25% — Passed 4/4 evaluators
Works with your favorite AI frameworks
From local development to production monitoring, Metahub gives you complete visibility into your AI applications.
Capture every LLM call, tool invocation, and agent step. Visualize execution flows with beautiful ASCII trees.
Evaluate completeness, latency, cost, and tool usage with heuristic scoring.
Detect prompt injection, data leakage, and OWASP LLM Top 10 vulnerabilities.
Find and redact emails, phone numbers, SSNs, and credit cards in traces.
Beautiful CLI output. JSON exports. Python SDK. Works offline, no account needed.
Export HTML/CSV/Markdown reports. Compare traces, benchmark performance, aggregate stats.
One pip command to get started. Works with Python 3.9+.
Trace any Python script that uses LLMs or AI agents.
Analyze, evaluate, and secure your traces instantly.
Automatically detect the most critical vulnerabilities affecting LLM applications.
Learn About Security🔒 Security Scan Results 🔴 CRITICAL prompt_injection.jailbreak Potential jailbreak detected 🟡 MEDIUM data_leakage.api_key OpenAI API Key in output 🔴 CRITICAL pii.ssn SSN detected in response ❌ 2 CRITICAL vulnerabilities found!
Get started with Metahub CLI in under 5 minutes. It's free and open source.