Plans and billing

Every workspace has one plan. It decides how much you can store, how many questions your assistants can answer, and which features are switched on.

The plans

Starter is one site with one assistant: a single knowledge base, a modest document allowance, and one widget. It is the plan for "we have a help page and we would like it to answer questions".

Pro is the working plan: several knowledge bases, every widget surface, custom CSS, the REST API and the connectors, and analytics you can export. Most teams that use WelcomeAI daily are on it.

Business adds room and control: many more knowledge bases and documents, white labelling, the audit log, and roles so people can be given only the part of the product they need.

Bring your own key is an add-on rather than a plan. Any of the three can switch it on: you keep the plan you have, and your own Anthropic and OpenAI keys answer every question, so those providers bill you at their prices and there is no monthly answer limit from us. It suits a workspace whose volume makes the model cost the biggest line, or one whose policy requires the provider relationship to be theirs. You need both keys -- Anthropic writes the answers and OpenAI makes your documents searchable -- and until they are stored the assistant will not answer at all. See the guide on using your own keys.

What a trial is

A new workspace starts on a fourteen-day Pro trial with no card. Nothing is charged and nothing is switched off during it. At the end you pick a plan; until you do, the workspace stops answering new questions but nothing is deleted.

How usage is counted

Two things are metered, and they are not the same thing:

Messages are answers your assistants gave. One question and its answer is one message. A visitor who asks six questions in one conversation spends six.

AI tokens are what those answers cost to produce. A token is roughly three quarters of a word, and both what we send the model (your question plus the retrieved passages) and what it writes back are counted. Output costs about five times input, which is why a long answer costs more than a long question.

Everything else -- documents, storage, knowledge bases, widgets, seats -- is a level rather than a meter. It is how much you are holding, not how much you have spent, so it goes down when you delete something.

Usage resets at the start of each billing period. The Billing and usage page shows every figure as used out of total, so you never have to work out how close you are.

At the limit

An answer already being written finishes. After that, new questions are refused until the period resets, and the widget says so rather than failing silently. Nothing is charged for going over -- there is no overage bill and no surprise invoice.

If one number is the problem and the rest of the plan fits, ask us to raise that one rather than moving up a whole plan. Raised limits show on your billing page with the original underneath, so you can always see what is yours and what was granted.

Buying through a link somebody sent you

An agency or a consultant may send you a link like app.welcomeai.dev/their-workspace/billing/c/6kk3eiuy7w82. It opens a payment page for one plan, with no sign-up form before it and no trial to start.

The price is the normal price. Whoever sent it earns a share of what WelcomeAI keeps, which changes nothing about what you pay or what the plan includes. Your email becomes your account, and the workspace is created when the payment confirms.

See Buying through a link someone sent you for what the page asks for and what the access question means.

Changing plan

Upgrading takes effect immediately and is prorated. Downgrading takes effect at the end of the current period, so you keep what you paid for.

Cancelling stops the renewal; the workspace keeps working until the period ends and then stops answering. Your data stays for the retention window, so cancelling is not deleting.

Refunds

A payment is not refunded once its period has started. Cancelling stops the next charge and the plan keeps working until the period you already paid for ends; the unused part of that period is not returned, and a downgrade is not prorated backwards.

The reason is where the money goes. An answer spends model tokens with Anthropic, a document spends embeddings with OpenAI, and an indexed passage occupies storage from the moment it is written. Those are charged to us as they happen and cannot be reclaimed, so refunding a period that has been used would return money that has already left.

This is why the trial is fourteen days of Pro with no card: nothing has to be bought to be evaluated. Index your real content, put a widget on a real page, ask the questions your customers ask. If it does not answer them, the trial ends and you have paid nothing -- and the workspace goes read only rather than being deleted, so nothing you built is lost either way.

What does go back, in full:

• a duplicate charge
• a charge taken after you cancelled
• an amount billed in error by us -- the wrong plan, the wrong currency, an add-on you did not buy
• an unauthorised charge
• the unused part of a period, if we end your agreement without cause

And nothing here removes a right you have by law. If you buy as a consumer and consumer law where you live gives you a right to withdraw or to a remedy, that right applies over this policy.

Write to billing@welcomeai.dev from the address on the account, with the workspace name and the date and amount of the charge. We reply within two working days. Please write before opening a dispute with your bank: a chargeback is decided on paperwork rather than facts, and while one is open we may suspend the workspace it relates to.

The full policy, including the statutory-rights detail, is at welcomeai.dev/legal/refunds.

Invoices and payment

Payment is handled by Lemon Squeezy, our merchant of record. They hold the card, produce the invoice and handle the tax; we never see a card number. Invoices appear on the Billing and usage page as they are issued, and the customer portal is where you change the card or the billing address.