● live demo — no install required

A real broken agent.
AgentLens finds the problem.

This is actual AgentLens output — the same UI you'd see in your terminal after running agentlens runs view.

// the scenario
🤖
Customer support agent — broken

An OpenAI agent with two tools: search_web for external queries and query_db for local customer records. A user asked it to find their renewal status. The agent called the wrong tool, hit a network error, retried once, and crashed. It never actually looked up the customer's record.

// agentlens runs view — embedded output
openai_customer_support_agent 070901b0 · 2026-06-06 21:23:44 UTC FAILED
4 spans 1 llm 2 tool 1 err 182 tok $0.000118 0.02ms
⚠ ROOT CAUSE: tool_selection · FAILED AT: Step 2 (search_web) · CONFIDENCE: 0.90
1 ● gpt-4.1-mini
0.02ms 182t
2 ▶ search_web
— ⚠ ROOT CAUSE
3 ▶ search_web
—
4 ✕ NetworkError
—

Try it with your own broken agent.

Two lines to instrument. One command to diagnose.

01 — install pip install runlens
02 — instrument import agentlens
agentlens.init()
03 — diagnose agentlens diagnose <run_id>
$ pip install runlens


★ Star on GitHub →