◉
Inspect agent runs.
Select a case file from the left.
What you'll see here
Every LLM call your code makes becomes a trace — the top-to-bottom story of one agent run. Inside it, each model call, tool call, or checkpoint you wrote in your code is a step. Click a step to see its evidence: what the model saw, what it returned, latency, tokens, errors. Mark any failing step as an annotation to build your regression set.
60 seconds to your first trace:
-
1
Scaffold a starter project
loupe init my-agent -
2
Set a provider key (Gemini is free — any of the six work)
export GEMINI_API_KEY=YOUR_KEY -
3
Run the agent — the trace lands here live
cd my-agent && python agent.py 'hello'
New traces appear here automatically — no refresh needed.
◉
This is the timeline of one agent run. Each row below is a
step — an LLM call, a tool call, or a
checkpoint you wrote in your code. Click any step to see
its evidence (inputs, outputs, HTTP call).