When an AI agent ships a bad answer, the question isn't 'what went wrong' — it's 'WHERE did it go wrong'. This replayer lets you step through any past run with full trace: prompts, retrieved context, model outputs, tool calls.
What's Included
- Full Trace Log: Every prompt, every retrieved context chunk, every model call, every tool call
- Step-Through UI: Click forward through each step like a debugger
- Diff Mode: Compare two runs side by side (good output vs bad output)
- Replay With Modification: Edit any step's input and re-run from that point
- Trace Search: Search across all past traces by content or tool invoked
How To Use
- Clone this app
- Connect to your Taskade agent via automation — traces start recording
- When an agent run goes wrong, find it in the replay log and step through
- Identify the broken step; modify and re-run