Skip to content

Sessions ​

A session is an agent working on one Issue in a Docker sandbox on one of your machines. The app also calls it a run. A session moves the Issue through seven phases: Intake, Ideate, Capture, Verify, Run, Validate and Close. On the way it claims REQs, ticks their criteria, archives them, and stops to ask you when it needs a decision.

Sessions list ​

Sessions lists the sessions in your projects. It opens on Live. The number next to Sessions in the sidebar counts live sessions, and a dot means one is waiting on you.

The Sessions list on Live: SHOP-003 in Capture, waiting on you with a question, started in a terminal; SHOP-002 in Run, holding REQ-005 at 2/4 while it runs tests, started on the web.

Each row shows what the session is doing now. The phase comes first, then either Waiting on you with the agent's question, or the REQ it holds with its criteria progress and the command it is running. Machine shows where it runs and whether it was started on the web or in a terminal. Signal shows how long ago the machine last reported on it. The dot is green under 30 seconds, amber up to 90 seconds, and grey once the session has gone stale or ended.

Filter with Search, Project and Status.

Past sessions ​

Set Status to All statuses to add stopped and failed sessions, or pick one status: Starting, Running, Stale, Stopping, Stopped or Failed. When nothing is live, the empty list tells you how many past sessions are under All statuses.

Sessions on All statuses: the two live sessions plus a failed SHOP-004 run on linux-builder and a stopped SHOP-001 run, both marked Ended.

Start a session ​

Creating an Issue in the app starts its first session. To start another, open the Issue and click Start session in Quick Actions. On On a machine, pick the inference provider, the model and the machine, then click Okay go. Each machine in the list says whether it is online and how many of its slots are free.

The Start session dialog on the On a machine tab: it will clone acme-example/storefront, with inference provider claude, model Claude Sonnet 4.6, and studio-mac online with 1 of 3 slots free.

The project needs a GitHub repository first, because the session clones it. Without one the dialog says so and points you to Edit details on the project. Okay go stays greyed out until you pick an online machine with a free slot.

From a terminal ​

The From a terminal tab gives you the command instead. Run dowork-runner start SHOP-004 inside a clone of the project's repo on any paired machine, or add --project with the project's slug anywhere else. The session appears in Sessions within a few seconds and your terminal follows it. Machines covers the command in full.

The From a terminal tab with the dowork-runner start SHOP-004 command, the same command with --project acme-storefront, and notes on following and detaching.

The session page ​

Open a session from Sessions, from Open session on its Issue, or from the dashboard. The header shows the Issue's title and code (click the code to go to the Issue), where and when the session started, its status and its signal. The phase stepper marks the current phase.

The live session for SHOP-002: header with Running, Signal 6s ago and Stop; the phase stepper on Run; the Work panel with REQ-004 archived and REQ-005 held; and the transcript with milestones and a message you sent.

The Work panel lists the Issue's REQs with their criteria progress. The REQ this session holds says Held by this session and opens to show its criteria, with Next up on the first one left. Above the transcript, Doing shows the command the agent is running right now.

The transcript shows the agent's messages, the tools it used, and anything you sent, labelled "You · web" or "You · terminal". Tracker milestones sit between them in time order: Claimed REQ-005, REQ-005 · AC 1 ticked with the evidence, Archived REQ-004. The runner's opening prompt is hidden because nobody typed it.

Message the agent ​

Type in Message the agent and click Send. If the agent is still working on your last message, Send queues the new one until it's ready, and Send now interrupts the current round instead.

The message box with a draft asking the agent to cover the empty state, the Send button and the Stop menu.

If the machine stops reporting for 90 seconds, sending turns off and the page says the machine hasn't checked in. It comes back when the machine does.

Answer a question ​

When the agent needs a decision it stops at a gate. The session page shows Waiting on you in the header and a Gate card with the question. Click one of the offered answers. When the agent offers no choices, type your answer and click Send answer. The session also shows as waiting on the dashboard, on the Issue and in Sessions, and you get a notification.

The SHOP-003 session waiting on you: a Gate card asking whether order history should include cancelled orders, with the answers Completed orders only and Include cancelled orders.

Your answer appears in the transcript as "Answered: …" and the agent carries on.

Stop a session ​

Click Stop for two choices. Finish REQ-005, then stop lets the session finish the REQ it holds and claim nothing new; it stops once that REQ is archived. The button then reads Stopping after REQ-005. Stop now kills the sandbox straight away, and unmerged work in its worktree is lost.

The Stop menu open: Finish REQ-005, then stop (claims nothing new, stops once this REQ is archived) and Stop now (kills the sandbox, unmerged work is lost).

The same menu sits next to Send under the transcript and on the Issue's live session strip.

Finished sessions ​

A stopped session keeps its page. The Work panel shows the REQs it archived, and the transcript keeps every message, answer and milestone, so you can see what the agent did and in what order.

The stopped SHOP-001 session: all three REQs archived in the Work panel, and a transcript of criteria ticks and archives ending with the agent's closing message.

Failed sessions ​

A failed session says why at the top, for example "The agent did not start. Retry the run." Retry starts a new session on the same Issue on an online machine, preferring the one it used last, and opens it.

The failed SHOP-004 session on linux-builder with the reason "The agent did not start. Retry the run.", a Retry button and the error in the transcript.