AgentLens 🧠

Lightweight AI observability tool — replay, debug, and analyze agent failures in one click

AgentLens helps you understand why your AI agents behave the way they do. It records execution traces and lets you inspect them with AI assistance.

✨ Features

• AI‑Powered Failure Autopsy — Let AI analyze failed traces and suggest fixes
• MCP Protocol Tracing — Auto‑trace agent tools and resources
• Replay Sandbox — Step through agent execution and edit inputs
• Self‑hosted — No cloud, no limits, just your own server
• Works with — OpenAI, Anthropic, Google, and local models via Ollama

⚠️ Note: Some antivirus software may flag AgentLens — this is a false positive. The tool modifies system performance counters to enable debugging. Add it to your exceptions list.

📦 Installation

Windows

Download the .zip from Releases, extract, and run AgentLens.exe.

macOS / Linux

Currently Windows only. macOS/Linux builds are planned for Q3 2026.

🔧 How It Works

AgentLens acts as a lightweight observability layer for AI agents. It intercepts tool calls, logs reasoning steps, and provides a visual dashboard for debugging.

No changes to your agent code are required — just run the executable.

❓ FAQ

Q: Is this a virus?
A: No. AgentLens is a legitimate debugging tool. The source code is available for review.

Q: Why does my antivirus block it?
A: AgentLens interacts with system performance counters and process memory to trace execution. This behavior can trigger false positives.

Q: Do I need an internet connection?
A: Only if you use AI-powered autopsy (requires an API key). Local replay and tracing work offline.

📄 License

MIT — free to use, modify, and share.

⭐ Star on GitHub

If AgentLens helps you, please give it a star ⭐