FRSelf-host it
Live demo — simulated traces from two agents. sentry-triage regressed when agent 1.5.0 and prompt v8 shipped; support-bot is healthy. Read-only.
Agents / support-bot / run_5dbf6290 / compare

Compare runs

B is compared with the previous run of the same agent.

Arun_a0b1ae89okv2.1.0
3 steps · 2.44s · $0.0009
Brun_5dbf6290okv2.1.0
3 steps · 3.20s · $0.0009

steps 3 → 3 · cost +0% · duration +31% · outcome succeeded

What changed between them

Same app version, agent version, prompt version and model.

Step by step

= same (2)≠ changed (1)+ only in B (0)− only in A (0)
A · run_a0b1ae89B · run_5dbf6290
=#1 route#1 route
≠#2 orders.lookup#2 orders.lookupargs changed
=#3 answer#3 answer

System spans are left out; steps are aligned on the tool called (or LLM call), then compared on arguments and outcome.