People live at stations. Cards visit them.
Focus on what you do best. The card comes to you. Own the station, not the ticket. People (and agents) live here; tickets don’t.
Why Stations?
Issue trackers treat work as a ticket someone owns. Chat tools treat work as a conversation. Stations treats work as the required input to an outcome.
A configured Stations board is a pipeline of stations shown as a kanban board. Cards move along that line. Each station asks for specific inputs and outputs, and some require a human to sign off before the card continues. A reject sends the card back; the board stays a line, which is why it reads as kanban even if you have never run an agent.
Built for agents as much as for humans.
Stations isn’t a human tool with a chat agent bolted on. You already have an agent and a harness; Stations meets them there instead of becoming one more harness to switch into. Humans and agents share the same claim, complete, and audit path — and the board stays a plain kanban board for the reader who only wants that.
FAQ
What is a station?
Professional kitchens divide the line into stations—grill, sauce, pastry—each responsible for a specific part of service. Cooks stay at their specialties while orders move through the kitchen, and nothing is served until every part is ready. Stations applies that model to project work. Each column defines who may act, what must arrive, what must be produced, and whether a human must approve it. Humans and agents work at the stations; cards travel through them. If the required outcome isn’t proven, the line stops.
Why isn’t there a Stations AI chat?
Context switching between tasks was hard enough; context switching between agents and conversations is a new headache. You already have an agent. Stations is the board that agent works — claim, complete, approve — through your harness. Chat can compile a plan. The board stays the source of truth. Actions live on the card and the station you are looking at, not in a default-home agent tab.
Where do humans and agents fit?
The worker is whoever claimed the station. Humans and agents are the same kind of actor, on the same claim, artifact, and complete path. Review is a flag on the station or a review station of its own, not a third assignee field.
Is this a ticket tracker with extra steps?
No. The board is a line of stations, not four hundred statuses. A card can’t advance until the required artifact — a pull request, a sign-off — is attached, and where a gate calls for it, a human approves. An agent claims a station with an expiring lease, the same as a human would. There is no Gantt, no sprint object, and no work OS bolted on the side. Find the card, claim it, complete it, approve or reject.
Isn’t this Step Functions, Temporal, or n8n with a board?
No orchestration engine to configure. Humans and agents work the same board through the same claim, complete, and approve path — a human-in-the-loop review is a station on the board, not a callback you wire up separately.
What will I see in the demo?
A board of stations. Cards with a claim. A ticket that can say, in a sentence, why this card is stuck. Human-in-the-loop as approve or reject, not a comment thread hoping someone notices.