01
Runs
Every agent run, readable.
Prompts, tool calls, guardrails, tokens and cost on one timeline. Open any run from a link and see exactly what the agent did.
THE PROBLEM
Logs report each attempt as its own line, so a single run is scattered across three streams. Finding the call that went wrong means reading all of them.
HOW TRACEWELL HANDLES IT
Each run becomes one trace: a tree of spans drawn to scale. Parallel tool calls share a row, slow spans stand out, and every span carries tokens and cost.
01
How it works
Three steps from nothing to useful.
01
Wrap your agent
One function call turns each invocation into a run.
02
Open the trace
See the waterfall first, then the words inside any span.
03
Share a link
Send a redacted run to a teammate or a customer, no account needed.
−8%
cost per run after the first week of traces
38 ms
added latency at p95
Try it on your own agent.
Free for 10,000 runs a month. No card, and the first trace takes five minutes.