ravel
Solana credits 12.50 Try a replay

A flight recorder for AI work

Rewind the run.
Keep what worked.

Record agent steps. Replay saved tool results. Branch from the mistake.

Try a replay How it works
  • Recorded inputs
  • Branch any step
  • Save test cases

Replay room

RUN #018 · Demo fixtures · change run

2 recorded tool results

Task

    Recorded run

    Branch from step 03

    Tool data

    Estimated rerun · 0.03 credits

    AI continuation uses credits.

    See what changed.

    Turn a mistake into a test.

    Open scenarios

    When the agent gets it wrong

    Read the docs
    1. Open the recorded runEvery model turn and tool call, in the order it happened.
    2. Change one instructionEdit it at the step where the run went wrong.
    3. Replay a new branchTool calls are answered from saved results, so nothing live is touched.
    4. Compare, then keep the casePut both answers side by side and save the run as a test.

    A debugger for AI agents and MCP tools · MVP: one built-in agent, tool-call recording and replay on the OpenRouter tool-calling loop