All workAI tooling
Tracer
Inspector for AI agent runs
When an AI agent takes hundreds of steps, Tracer lays them out on a timeline: what ran in parallel, where the time went, what the model got and returned, and where a tool failed.
- Type
- Product
- Year
- 2026
- Role
- Product design and frontend
- Stack
- Next.js, React, TypeScript
- Links
- Try the demo
Challenge
A developer chasing a bug in an agent run needs to find the step that went wrong, fast.
Approach
Failed steps are marked in red, and a single key jumps between them. Each step shows its messages, JSON, tokens and cost, and you can drop in your own trace file. The data never leaves the browser.
What we built
- Step timeline with time totals by type
- Messages, JSON and errors for every step
- Tokens and cost per step and per run
- Keyboard navigation and file import
