Skip to content

Cost model

Controller AI meters usage in usage cents (100 usage cents = $1) drawn from a pool shared by the whole organization. Every run — development or live — draws from the same pool. This page gives you the shape of the model so you can predict a workflow’s cost before running it at volume; exact rates are visible in the app under usage records.

WhatCost shape
Flow control, logic, expressions, data records, agent file operationsFree
Built-in integration nodesSmall flat fee per execution (smaller when using your own connected account)
Catalog integration nodesFlat fee per execution
AI operations (Use AI, AI-powered expression ops)Provider’s cost with a small markup
Paid external APIs proxied through the platformProvider’s cost with a small markup

Two consequences worth internalizing:

  • Structure is free. Branching, transforming, joining, storing records, and returning responses cost nothing. You never pay for making a workflow well-organized.
  • Volume multiplies the metered steps only. A flow run over a 500-item list pays 500× its integration and AI steps, and 0× its logic. When estimating, count the metered nodes inside the loop, not the nodes in the workflow.

Each workflow execution records its cost, visible in execution history and in usage records under settings. Usage records break down by category: AI provider usage, managed integration usage, external API usage, integration action calls, and execution costs.

Terminal window
cai exec list
cai exec get <workflowExecutionId> --json

Agent conversations meter their tool calls the same way — a workflow tool’s cost shows on the workflow execution it creates.

What your subscription covers — and doesn’t

Section titled “What your subscription covers — and doesn’t”

Your coding-agent subscription (Claude, Codex, Cursor) covers the reasoning of the agent that builds on Controller AI. It does not cover the Controller AI hosted runtime: agent conversations on the platform, workflow runs, and AI nodes bill to the organization’s usage pool.

  • Prefer data-store records over repeated external lookups — reads and writes are free.
  • Put the metered step behind the filter, not before it: filter the list first, then run the integration node on survivors.
  • For scheduled workflows, remember the schedule multiplies everything downstream of it.
  • Before a bulk run, check the per-item cost with a single dev run and multiply.
Estimate what one run of this workflow costs, broken down by node, and what a daily schedule over ~200 items would cost per month.
Restructure this flow so the AI step only runs on records that pass the filter.