Runs and allowances
Every plan is a fixed price for a fixed allowance. Nothing is metered: the workspace never shows you a bill for model tokens or engine answers, and nothing costs more because a site is large.
What the allowance is
Section titled “What the allowance is”Two parts, both per site:
- The poll: the topics a site measures and the AI answers the poll records a month. The answers are spread through the month, so the poll never runs out early. See The poll.
- The site audit: full runs and change runs a month, and the pages each run reads.
| Starter | Growth | Portfolio | Enterprise | |
|---|---|---|---|---|
| Full runs per site a month | – | 1 | 1 | 1 |
| Change runs per site a month | 1 | 4 | 4 | 8 |
| Pages crawled per run | 1,000 | 1,000 | 2,500 | 5,000 |
| Pages analysed in depth per run | – | 300 | 400 | 600 |
Topics and poll answers per plan are on Plans and limits.
Of the pages a full run crawls, the model analyses a share in depth, one of every template first; the rest get the access and technical checks. A Starter site gets one baseline crawl when it is verified, read without the model and outside the allowance, so its change run has something to compare with.
Allowances reset monthly on the billing day, the day of the month the plan started, on annual plans too. Settings, Usage shows what each site has used, and the run dialog says when the next reset is.
Full run and change run
Section titled “Full run and change run”A full run samples up to the page budget across every template and language, fetches each page as every machine identity, builds the knowledge graph, simulates retrieval and writes the analysis. Its findings feed the page plays and Foundations.
A change run re-fetches exactly the pages the last full run sampled, so every difference is a change on the site and not a change of sample, plus the newest URLs the sitemap has listed since. It re-checks access, technical facts and performance, and reports what moved: new, resolved and regressed findings, pages whose content changed or stopped returning 200, and pages that AI crawlers can newly read or are newly refused. It does not rebuild the knowledge graph or re-run the analysis; the full run’s findings are carried into it, labelled as such. It is the regular check that nothing broke, and the run to start after a template ships.
Runs can be started by hand or by the site’s schedule; both count the same.
Pages written for plays
Section titled “Pages written for plays”The page Vercatus writes for a play is the one thing counted by the workspace, not the site: Starter 6, Growth 30, Portfolio 120 and Enterprise 400 a month, as the contract sets. Each page that is delivered counts one, a new version of the same page too; a page that fails its checks, a play’s steps and a play’s short drafts count nothing. At the limit the request is refused with the reason and the allowance’s renewal date; Settings, Usage shows what is left.
What is not counted
Section titled “What is not counted”The daily access watch, acceptance checks, source reading, search lookups, the asks without search and the Engine Briefs are part of the plan and spend nothing you can see. They are budgeted on our side, separately from the poll, so they never stop it.
Before you start a run
Section titled “Before you start a run”The run dialog, vercatus estimate and POST /api/estimate all say the same thing before anything starts: which
allowance the run consumes, how much is left and when it resets, how many requests it makes, and roughly how long the
crawl takes.
At the limit
Section titled “At the limit”A run that would exceed the site’s allowance is refused with one sentence (402 from the API) and nothing is started;
the next run of that kind waits for the reset. A topic beyond the plan’s topics is refused with the reason. A site
beyond the plan’s included ones is priced per site and prorated the moment it is added; a plan with no room for another
site says so, and Settings, Plan switches to the larger one in place. Deeper runs, with more pages per run, are on
request. Nothing is added to an invoice without your action.
Your own model provider
Section titled “Your own model provider”An Enterprise workspace can route its model calls through its own Anthropic API key, Amazon Bedrock account (an access key, or a role Vercatus assumes with your workspace id as the external id) or Google Vertex AI project (a service account key). An admin sets it under Settings, Model provider, and tests the connection there before or after saving. From the next call, the workspace’s model calls go to your provider, which bills you; that provider is then no longer our sub-processor for your workspace. The poll keeps asking Claude itself, since it measures what that product answers. Tokens still count against each run’s budget; Settings, Usage shows them with an estimate at list price. Every other workspace runs on the platform’s providers under no-training terms, and model spend is part of the plan price.