• Pricing
Sign inStart free
Start free
Contents
  • The definition
  • Token budget per action
  • What does not count
  • Multi-task requests
  • Compound actions
  • Requests that read many records
  • Voice
  • Agents
  • Document templates
  • Monthly quotas
  • You won't get a surprise bill

What counts as an AI action

Ortely's AI is bundled into your subscription, not sold separately. So that your bill is predictable on both sides, every AI feature consumes one published unit: an AI action. This page is the canonical definition — the same text referenced by the Terms of Service and shown in-product on your usage meter, so you can sanity-check your monthly bill against your actual day-to-day work. An opaque token meter would feel like a hidden tax; this is the opposite of that.


The definition

One AI action is one discrete, platform-initiated task that produces a single, defined output:

  • One document extracted (PDF or image → structured fields) = 1 action
  • One Cmd-K assistant question answered = 1 action (more when it reads many records — see Requests that read many records below)
  • One carrier rate sheet AI-imported = 1 action
  • One quote auto-drafted = 1 action
  • One inbound email AI-classified = 1 action
  • Five minutes of bundled voice transcription = 1 action
  • One document template built from your document = 5 actions; each chat change to a template = 3 actions

AI work counts when the AI does it, whether or not you then create, save or confirm anything. Reading a document, each assistant or voice turn and each regeneration are actions, and they count toward your included actions and any overage.

Building a custom document template with AI, from a document of your own, uses more actions than most tasks, and so does each change you ask for in the template chat, whether it starts from your document, your saved template or the built-in one. You keep one custom template per document type; saving a new one replaces the previous one.

Token budget per action

Each action carries a default budget of roughly 5,000 input tokens and 2,000 output tokens — about ten input pages for a document extraction. A task whose prompt or response exceeds that budget counts as additional actions in proportion to the tokens used, rounded up. A 30-page rate sheet is three actions; a 50-page rate sheet is five. The budget is generous for normal work — it only matters for unusually large inputs.

What does not count

  • Manual edits you type don't count. Only AI-initiated work does.
  • Failed-and-retried calls don't double-count. A retry after an error is still one action.
  • Cached-prompt hits do count. The meter is on the user-perceived task, not the per-token cost — so it tracks what you actually asked for.

Multi-task requests

Combining several discrete tasks into one request does not make them one action. The system detects multi-task prompts — a single Cmd-K query that asks the assistant to "list X, then draft Y, then summarize Z, then create W" — and either splits them into separate actions or asks you to split them yourself. This rule is published here and in the Terms of Service so it is visible before you agree to it.

Example — a true single action

You open Cmd-K and ask: "Show me shipments to Brazil this month where cargo is electronics." The assistant returns a filtered list. One action.

Example — a multi-task request that counts as several actions

You open Cmd-K and ask: "List all my shipments to Brazil, then draft fifty customer-update emails about delivery status, then summarize my month's revenue, then create an invoice for customer X." That is four discrete tasks bundled, producing fifty downstream email drafts on top of three retrieval and reasoning steps, and token usage roughly an order of magnitude over budget. The system either asks you to split it into separate actions or, if you confirm you want to proceed, counts the request as approximately ten actions.

Compound actions

Some requests create several linked objects at once — "set up a customer on Net 30 terms" creates the entity and its payment terms; "book a shipment with three boxes" creates the shipment and three pieces. A compound action counts as the number of objects it persists: the entity plus its payment terms is two actions; a shipment plus three pieces is four. This is the same per-output rule as above, applied to one confirmed request instead of several bundled ones. Roles and tags that identify an entity don't add to the count. The extra actions for the objects are counted only when you confirm — discarding a draft adds nothing beyond the request that drafted it.

Example — one request, seven objects

You ask the command bar or voice assistant to "create a customer with three addresses and three contacts." That is one request, but it persists seven objects: the customer, its three addresses, and its three contacts. It counts as seven actions — one per object created. If you discard the draft instead of confirming it, nothing is created and only the request that drafted it counts.

Requests that read many records

A question that looks up a handful of records is one action. One that reads many counts by what it read: each kind of lookup the assistant does for you counts one action per ten records it returns, rounded up. Showing you the records is what counts, not how many exist — the same list on the page costs nothing. Finding the record you named (a lookup that returns that one record), opening a page and searching help don't count.

Example — a list, then a figure for each

You ask: "Show me last week's shipments, and the profit margin for each." The assistant reads 40 shipments (four actions) and their 40 margins (four actions): eight actions. Had there been 8 shipments, it would be two actions — one for the list, one for the margins.

Creating or changing records still counts one per record, as above.

Voice

The voice assistant is metered two ways, and they don't overlap:

  • Talking to it — the spoken conversation and transcription — draws on your plan's bundled voice time, billed in five-minute blocks (five minutes = one action). Your remaining voice time is shown on the same in-product usage meter as your actions.
  • What it does for you — every task it performs (answering a question, drafting an entity, composing a quote) is an AI action, metered exactly like the typed command bar, including the compound rule above. Speaking a request instead of typing it doesn't change what it counts as.

As everywhere else, the assistant drafts; nothing is created or charged as a compound action until you confirm.


Agents

Agents are background workers you switch on in Settings, on Growth and above. Each one looks for one kind of stalled work and reports it, drafts the next step for you to approve, or, on Pro and above and only where you allow it, does it on its own.

  • A run that only checks your records and calls no model is free.
  • A run that drafts or acts = 1 action, or more under the token budget above, exactly as if you had asked the assistant yourself.
  • Agents spend the same monthly quota as everything else. Past it, agent actions are billed at the same $0.30 per action, itemized on their own invoice line.
  • Every agent has a monthly budget you can change. You are notified at 80%, and the agent pauses at 100% until the next period.
  • Agents pause on their own when your overall AI usage reaches 150% of your included actions, and you can pause all of them at once.

Document templates

In Settings › Document templates you can ask the AI, in a chat, to change the layout of one of your printed documents, starting from the layout it prints with today. Or you can upload a document you use today and have the AI build a template like it, then adjust it in the same chat. Reading your document and writing a whole page layout is more work than an ordinary task, so these steps count more:

  • Building a template from your uploaded document = 5 actions.
  • Each chat change to it = 3 actions, and so does each automatic fix of a change the template checks refused.
  • They count whether or not you save the template.
  • Previews, saving, switching a template on or off and removing it count nothing.

Monthly quotas

The quota scales with your tier. Shipment volume is never capped on a paid plan — the AI-action quota is the meter that scales with usage, and it is shown in-product so you can read it at any time.

How many actions each plan includes, and the overage rate past them, are published in one place: the AI actions row of the plan comparison.

You won't get a surprise bill

Any usage past your monthly included actions is billed at the per-action overage rate published at ortely.com/pricing. If usage keeps climbing, the AI layer pauses gracefully with a notification and a one-click upgrade path. Manual workflows are never affected.

Separately, a few hard limits refuse a request before any model is called: file uploads above 25 MB, single documents above 50 pages, sustained traffic above 60 actions per minute, and a per-tier daily ceiling. These exist to stop abuse and accidents, not normal use.


Questions about your usage or your bill? Email support@ortely.com.

How many actions does each plan include?

Compare plans and included AI actions →
Ortely

AI-first freight software for forwarders, freight brokers, cargo GSAs and logistics teams.

hello@ortely.com
Product
Product overviewWhat our AI doesAgentsSOPsPricing
Solutions
For freight forwardersFor freight brokersFor cargo GSAsFor logistics teamsFor air carriersIntegrations
Resources
Help centerGet startedWhat's newBlogSecurity
Company
AboutPartner with usAccessibilitySitemap
Legal
DPACookiesAcceptable useSub-processorsResponsible disclosureAI actions
© 2026 Ortely. All rights reserved.
Terms|Privacy
Signing you in…