A run, turn by turn

Open any agent from its card and you land on its page. The default view is a summary: a few measured facts (when it started, what it’s delivered, what needs you), then one row per card it’s working, each with the agent’s own line about where that card stands. If it asked a question, you answer it right on that row.

Read the run

At the foot of that summary is Read the run. It opens the same agent’s work laid out turn by turn, in order, grouped by the stretches of time it was actually running. Each turn is your prompt (or the previous card handed over) and the agent’s answer to it, so you can read the whole conversation exactly as it happened rather than a rewritten account of it.

Open a turn and you get its trace — the CLI’s own stream as it ran: what it said, what it thought through, and every tool call it made. Nothing here is summarized or interpreted; it’s the CLI’s own output, shown as it was. A long trace shows its tail with the rest one click away, and it says how many earlier steps aren’t shown.

Commits and diffs

When a turn delivers, its commit shows up as a sha you can open. On a wide screen it opens beside the run in a pane; on a narrow one it opens as a sheet. Either way you’re looking at the real diff for that commit, tied to the turn that produced it.

Tokens spent

The agent page shows what the run has consumed in tokens — a figure like “12k tokens”, with the breakdown between input, output and cache available on hover. This is a token count, not a cost: Flowviant never shows a dollar figure for a run.

See When an agent stops for what a stuck turn looks like, and Review and merge for what happens once an agent has something to review.