Plans & usage

Three limits — runs at once, runs a month, seats — and what happens when you reach each.

Plans meter runs and seats, never features. Every plan includes all the trackers and notifiers, gated workflows, the tracker, the wiki, and the live session view.

The catalog

How the limits behave

They fail differently, on purpose.

• Concurrency is a queue, not an error. A run started past the cap is created as pending with a note explaining why, and starts automatically the moment a slot frees. The cap counts runs doing real work — running, waiting on an answer, holding a preview, or warm in an interactive step. A run parked at a gate holds nothing and does not count.
• The monthly quota is a hard ceiling. It counts runs created this calendar month, UTC, resetting on the 1st, across the whole organization. When it is spent, new runs are rejected with a message naming your plan and the reset date.
• The seat cap is checked when somebody joins. It counts the people already in the organization plus invitations still outstanding, because an invite link stays redeemable for a week — that is a seat somebody is holding. At the cap, inviting is refused, and so is redeeming a link that was sent while there was still room. Removing a member or withdrawing a pending invitation frees a seat immediately.

Watching usage

Settings → Billing shows your plan, live meters — runs this month, AI tokens this month, active runs, members against your seat cap — and the full catalog, counted by the same code that enforces the limits. The dashboard overview repeats the usage strip, and the apps show the same figures under Settings.

AI tokens are the model usage Peractor pays for: phases that run on Peractor's machines, the assistant, and prompt improvement. Every plan includes a monthly allowance shared across all three, counted per organization.

Runs on a machine you paired use that machine's own credentials and are never counted against it.