Which of your agent's calls went wrong?
Drop an export and see the calls where your agent went silent, looped or kept hitting a failing tool, and what to fix first. Free.
It runs in your browser. Your file is not uploaded. The same detector our command-line tool uses is loaded into this tab and reads the file here. Open your browser's network tab and watch: after the page has loaded, checking a file makes no request.
No export to hand?
What happens to your file
- When the page opens it downloads two things: a Python runtime (Pyodide 314.0.7, from cdn.jsdelivr.net, pinned to that version's folder) and our checker (from this site). That is the last network activity.
- Once both are loaded, the page turns off its own network functions and tells the browser to block connections from it.
- Your file is read into this tab's memory, checked, and dropped. It is not written to the address bar, to cookies or to browser storage. Closing the tab clears it.
- There is no analytics script and no sign-up on this page.
What it checks, and what it doesn't
It finds execution failures: tools that errored or timed out, slow replies, loops and dead air. It does not judge whether a business task was done correctly, and it can only see what your export recorded.
We are an early company. This checker has been tested on synthetic exports, a public benchmark and real public exports from Langfuse, LangSmith and OTLP GenAI (Strands, Vercel AI SDK, Mastra), not yet on a customer's production traces. The Vapi and Retell readers are checked against each vendor's current API spec; the OpenAI Agents SDK and Pipecat readers follow the documented format. None of those four has been run on a real export yet. If it misreads your export, we would like to hear which format it was.