Diagnose your own agent trace. Paste or upload a recorded run trace — a single run object or a JSON array of runs (id, agent_id, status, steps[…]). The server classifies each run's failure mode, pins the root-cause step, and returns a fix — computed entirely from your trace. It analyzes a recorded export, not a live agent. Nothing is stored.

Choose JSON Download sample

Guided tour

60-second walkthrough

Bring a run trace

Paste or upload a recorded agent run with steps, status, token counts, latency, errors, and model parameters.
Classify failure server-side

Analyze your trace sends the run to the backend, which identifies the failure mode, root-cause step, and suggested repair.
Inspect the black box

Timeline, tool tree, prompt diff, and token readouts make the failed run reviewable without rerunning the agent.

Open Tracewell proof repo

Diagnosing runs…

Run Timeline loading…

—

Select a run to view trace.

Select a step to inspect.

What this is

Portfolio demo showing production agent observability concepts
50 synthetic runs across 3 fictional agents at Redwood Labs (B2B SaaS)
Token counts derived from real prompt lengths — not fabricated
The v3→v4 compliance policy regression models a real class of production bug
Failure modes match what real teams get paged for at 3 AM
“Analyze your trace” runs a real Cloudflare backend — server-side rules diagnose a trace you paste; the default 50 runs are synthetic

Honest limits

No live model — replay routes to a local JSON fixture, not a real LLM
No real agent data — all runs, companies, and contracts are synthetic
No ingestion SDK — this does not instrument real agents; it visualizes pre-recorded data
No persistent storage — session state is in-memory only; refreshing resets everything
No multi-agent orchestration graphs — single-agent linear step trees only
Prompt diffs compare system-prompt snapshots, not full multi-turn conversation histories

60-second walkthrough

Bring a run trace

Classify failure server-side

Inspect the black box

Diagnosing runs…

What this is

Honest limits