An agent slot is the unit of parallelism. One slot runs one task at a time. How many slots you hold determines how many tasks can be in progress simultaneously.
The thing everyone gets wrong first
Slots do not make a task faster.
A task's duration is set by the size of the change and how long your build and test suite take. Buying a second slot does not halve that. It lets a second task run alongside the first.
If one task is slow, look at the test suite and the scope of the change. If your queue is long, look at slots.
How work queues
A task you create enters PENDING and waits for a free slot. When one frees up,
the task moves to IN_PROGRESS.
With one slot and five tasks, you get one at a time in order. With five slots and five independent tasks, all five run at once — assuming the work genuinely is independent. A plan whose phases depend on each other serialises regardless of how many slots are idle, because step three cannot start before step two exists.
Priority
Tasks carry a priority from P0 (highest) to P3. When more tasks are waiting
than there are slots, priority decides who goes first.
Priority orders the queue. It does not add capacity — if everything is P0,
nothing is.
What a slot costs
Slots are the main axis of the pricing model:
- Free — one low-priority slot, one project, capped agent-hours per month, and a ceiling on how long a single task may run.
- Professional — one standard slot included; additional slots are billed per slot per month. Unlimited projects and members.
- Enterprise — contracted slots with a minimum commitment, plus priority processing.
Model usage is billed separately from slots — see Model credits. A slot buys you a place to run work; the tokens that work consumes are their own line.
Sizing your capacity
Start with one slot and watch the queue for a week.
Add slots when tasks are consistently sitting in PENDING while you wait. Do not
add slots because individual tasks feel slow — that is a different problem with a
different fix.
Agent-hours
On plans with an agent-hour allowance, time counts while an agent is actively
working a task — not while the task sits in PENDING. Queued work is free;
running work is what consumes the allowance.