reviewmy / Products / Agent dashboard / Live demo

Three agents, one ticket, forty-one seconds.

A real run, replayed. Who did what, in what order, what it cost, and the moment it went sideways, then sent back as a review request without leaving the page.

run-4f1cstar-mountain/siteRunning0:00$0.05
0s15s30s45s60s
Plannerclaude-opus
Read ticket & scope
Coderclaude-sonnet
Read Cart.tsxEdit Cart.tsx · +18 −4run_tests · retried 2×Edit CouponField.tsx · +64
Reviewerclaude-haiku
Review diff · 4 filesOpen PR #218
Plannerreading the ticket
CoderWaiting,
ReviewerWaiting,

Now

Planner is reading the ticket

ticket #418

#418, "Coupon field accepts expired codes."
Scoping the files that touch applyCoupon().

Audit log

00sPlannerRead ticket & scope6s
06sPlannerWrite plan · 4 steps3s
10sCoderRead Cart.tsx6s
16sCoderEdit Cart.tsx · +18 −48s
24sCoderflaggedrun_tests · retried 2×17s
41sCoderEdit CouponField.tsx · +648s
49sReviewerReview diff · 4 files7s
56sReviewerOpen PR #2184s
0files written
1tool calls
0flagged steps
$0.05run cost

What you are looking at

Three views of the same run.

01

Lanes

One row per agent, laid on a shared clock. Parallel work reads as parallel, and a step that ran long is as wide as it felt.

02

Handoffs

The graph shows who holds the work right now. When the Planner finishes, the Coder lights up. Nothing is inferred from the log after the fact.

03

Evidence

Every line in the audit log names the agent, the tool, the arguments and the duration. The diff beside it is what actually landed on disk.

  • Replay, not a recording

    Scrub the run and the graph, the log and the diff all move to that moment. Nothing is a video.

  • Cost per step

    Spend accrues against the step that caused it, so a retry loop is visible as money, not just as time.

  • Flag and send back

    Anything that looks wrong becomes a review request the agent can pick up over MCP, with the step attached.

Watch your own agents work.

Connect an environment and the next run shows up here.