Try spwn in your browser

This is a simulation. It runs entirely in this page: no agent, no repository and no spend. The sessions, replies and costs are canned, but the layout and the flow are the real ones.

Simulated demo — no real agent, repo or spend↺ Reset

Things to try

1. Answer the blocked agent. “Fix flaky checkout test” is waiting on a permission prompt at the top of the Fleet. Click an option, or press 1. It goes back to work, then shows up under Ready for review.
2. Land two sessions at once. Press Land on two ready sessions. They go through the land order one at a time: the second waits as Queued #2 while the first syncs, checks and merges.
3. Fork an attempt. Press ⑂ on a session, pick how many, and give the forks a first message. Each fork gets its own branch and starts with its parent’s conversation. Its cost counts only what it spends itself, so the Fleet total never counts the shared history twice.
4. Land a fork. A fork’s base is its parent’s branch, so landing one merges it into the parent, not main. Watch the parent’s ahead count go up.
5. Talk to a session. Click a session to open it, type a message and press Enter. Each finished turn is a commit: the ahead chip and the cost both move.
6. Start a new session from the sidebar. It gets a fresh worktree and branch, created by the session-created hook.

What the demo leaves out

The running total in the Fleet header is the demo’s own addition; spwn shows cost per session. The real agent pane is the agent’s own terminal UI, not a transcript. The demo also has no Inspector (diff, timeline, PR doc), no Compare view, no pull requests, no editor and no conflicts: every sync here is clean. To see those, install spwn.