Reproduce the bug your customer found.
A conversation goes wrong in production and the bug report is a screenshot. Voxli imports the conversation and replays it until the failure reproduces, then keeps it as a test so the same bug cannot ship twice.
A customer conversation goes wrong in production.
No exception fired, so there's nothing to step through. The failure is probabilistic: the same conversation can pass nine times and fail on the tenth. Ammar Khan of Expertise.ai described debugging that way: "Engineers were putting in the same question again and again, hoping the issue would show up... It was absolute insanity."
Replay it until it shows itself.
Voxli re-runs the imported conversation until the failure appears, with every turn and tool call recorded.
From live issue to permanent test.
Import the conversation, reproduce the failure, and fix it without leaving your editor.
-
Import and replay the real conversation
Bring the conversation in from your logs or through the API and re-run it against your agent, exactly as the customer had it. The person who found the issue sends you a run link instead of a screenshot.
-
Repetitions corner the flaky failure
Run the same conversation up to ten times. A failure that only appears in some runs stops being an anecdote and becomes a rate you can work on.
-
Fix it from your editor
Connect Claude Code, Cursor, or any MCP client. Your coding assistant reads the failing run turn by turn, applies a fix, and runs the test again.
“The feedback loop is super fast... that would have taken hours before. Now it just kind of goes.”
Bring us the conversation that went wrong.
We are early stage and founder led. Book a call and talk directly with the people building Voxli.