Cost & limits

Cost & limits (Administration ▸ Access & Billing) answers three questions for owners and admins, one per tab: what am I spending, what will stop me, and where is it going.

1. 1

   Overview: what this month has cost

   The month’s total, the number of caps currently in force, and how many things are within 20% of one. The figures come from the same ledger your invoice is built from, so this page and your bill cannot disagree. If usage billing is not switched on for your workspace it says so: “used” and “owed” are different sentences.

2. 2

   Limits: what will stop you, and what enforces it

   Every ceiling the server can name, grouped by area (workspace, phone and SMS, email, AI agents, automations and the API), each with its current value, what actually enforces it, and where it is changed. A ceiling marked “shared” is a platform-wide one held in common with every other workspace, not yours alone. The four ceilings that stop an automation (webhook calls a day, actions in one run, steps in a chain, branches from one point) are listed here as well as on the Automations page, because a rule that goes quiet at teatime looks broken rather than capped.

3. 3

   Spend detail: line by line

   Every metered item this month with its quantity, unit rate and amount, adding up to the total on the Overview tab. These are the exact lines an invoice is built from.

4. 4

   Automatic alerts

   When a capped resource crosses 80%, 90% or 100%, owners and admins get an in-app (and, if email is configured, emailed) alert, once per band, so you are warned before you hit the wall rather than spammed.

5. 5

   Cap AI agents

   Set a maximum number of runs, tokens, or dollars, per day or per month, and agent runs stop before the model is called when any of them is reached. All three legs are real: every agent run records its token count and its cost, priced from the model that actually ran, so a dollar cap fires the same way a run cap does. (Until v1.170.0 the dollar leg was stored but never fed - runs reported tokens and zero dollars - and this page said so. The warning retires itself on the first priced run after the upgrade, not before.) Failed and cancelled runs hand their run-count back - you are never billed a per-run fee for a run our provider dropped - though the tokens they really consumed stay on the meter. Chief chat, enrichment sweeps and background watches are not agent runs and are not on this ledger.

6. 6

   Workspace AI budget (everything else AI)

   Separate from agent-run caps, a platform admin can give each workspace a daily AI budget in dollars, or one platform-wide default, under Administration ▸ API keys ▸ AI model router. It covers every on-request AI helper: Chief chat, briefs and setup blueprints, the writing assistant, the agent builder, agent proposals and enrichment, meeting minutes, AI fields, the document composer and the docs help assistant. Quill’s whole-sentence correction spends against the same budget and is the one thing here that runs without anybody pressing a button, so a workspace that wants nothing spent unprompted switches it off on the Quill page. At the budget those features pause politely until the daily reset, and anything that needs no AI call keeps working. Meeting and call transcription meters its spend into the same budget but never stops a recording already made. No budget is set by default, so nothing changes until one is set; a budget of zero pauses a workspace’s AI deliberately.

7. 7

   Operators: platform totals

   Platform admins see an extra “Platform totals, all workspaces” panel with cross-tenant email/SMS/agent spend for the month, so runaway cost across the fleet is visible at a glance. It is labelled as our cost, not any customer’s.