How message credits work

What it is

A message credit is what your chatbots' answers are paid in. Every time a chatbot replies to somebody, credits are spent from your workspace. How many depends on the chatbot's AI model: a 1X model spends one credit per reply, a 12X model twelve. The default Agentency AI model is 1X, so on it one answer is one credit.

That is the whole rule, and it is worth stating the corollaries plainly because each of them surprises somebody:

• The visitor's message is free. The reply costs. On a 1X model, a conversation of ten exchanges costs ten credits, not twenty.
• Every surface spends from the same pool. The website widget, the hosted page, a connected messaging app, and your own test chat drawer all draw on the owner's credits. There is no separate testing allowance, and dashboard tests count the same way — at the model's multiplier.
• Credits belong to the account owner, not to the person chatting. A team member's work spends the owner's credits, and always did.
• A cached answer still costs its model's credits. The visitor received an answer, so they were charged for one.
• Feedback and lead forms are free. A thumbs-down and a completed contact form cost nothing.

You have two pools, and they are consumed in a fixed order: your plan's allowance for the current period first, then any purchased pack balance underneath it.

Model multipliers

Every AI model is shown with its multiplier, from 1X to 12X. The multiplier is the number of credits one reply costs: 1X is one credit per reply, 3X is three, 12X is twelve. The default Agentency AI model is 1X.

• On the Free plan every chatbot answers with the Agentency AI model, so every reply costs exactly one credit.
• On a paid plan you choose the model per chatbot: open the chatbot's Playground tab and select Change AI model on the AI model card. The picker lists every model with its multiplier. Pick a live model, which answers visitors everywhere, and optionally switch on Test with a different model to give your playground and Test Chat a separate testing model.
• Visitor replies are charged at the live model's multiplier; your tests at the testing model's. Without a separate testing model, tests use — and cost the same as — the live model.
• A conversation keeps its model. A new pick applies to new conversations. One that already has messages keeps answering, and charging, at the model it started with.
• Downgrading to Free puts every chatbot back on the Agentency AI model at 1X. Your saved picks return if you upgrade again.

If credits are tight, a 1X model stretches your allowance furthest.

When you would use it

Before a campaign that will multiply your traffic. When chats stop working for everybody at once. When you are choosing a plan and want to know which number actually matters — for most accounts, credits are the limit that runs out first.

Where to find it

Dashboard → Settings → Billing shows the full meter. Owner-only.

The sidebar also carries a compact credits indicator, visible without opening Billing.

Your plan's allowance is on the pricing page and on your own Billing page, both of which read the live catalogue.

Steps

1. Open Dashboard → Settings → Billing as the account owner.
2. Read the credits meter. It shows what you have used against your plan's allowance for this period.
3. Look underneath for your add-on balance. Purchased credits appear separately and behave differently — see below.
4. Work out your burn rate: credits used divided by days elapsed in this period, multiplied by the days remaining. That tells you whether you will make it to the reset. Remember that a chatbot on an NX model spends N credits per reply.
5. If the answer is no, decide between a one-time pack and a plan upgrade — see Buy more credits.
6. If usage looks wrong, check whether your own testing is a large share of it. Test chat is metered exactly like real traffic.

When your allowance resets

On a paid plan, the reset follows your billing cycle, not the calendar. If you are billed on the 15th, your credits refill on the 15th. Do not plan around the 1st of the month unless that happens to be your renewal date.

Unused plan credits do not roll over. Whatever is left when the period rolls is gone. This is the standard arrangement for a metered allowance and it is worth knowing before you buy a large tier "to build up a buffer" — you cannot.

Purchased pack credits do not reset and do not expire. They sit underneath your plan allowance indefinitely until they are spent. The Billing page states it directly: "Monthly message credits reset at the start of each billing period. Purchased add-on credits never expire."

One important exception to know about: changing plan mid-period does not refill your allowance. An upgrade raises the ceiling for the next period and gives you the new plan's other entitlements immediately, but the credits you have already used this period still count against you. If you upgrade specifically because you ran out of credits today, buy a pack as well — see Buy more credits.

What running out looks like

There is no grace period and no soft overage. At zero, the chatbot stops answering.

Your visitors see a neutral message. On the hosted page it reads "The assistant is temporarily unavailable. Please try again later." In the website widget they get your configured fallback message, or a generic apology if you have not set one. On a messaging app they get "I'm temporarily unavailable. Please try again later." None of them are told anything about your account or your bill.

You see the real reason. In test chat, the message names billing explicitly and offers a link straight to Billing.

Nobody emails you. There is no low-credit email, no exhausted-credits email, and no notification. The only warnings are in the product: the sidebar indicator, and a banner on Billing once you drop below a fifth of your allowance reading "You're running low on message credits. Top up to avoid interruptions."

That absence is the single most important thing in this article. If credits matter to your business, put a reminder in your own calendar to check the meter. Do not wait to be told.

Controlling what you spend

• Turn the chatbot off where it is not earning. A paused chatbot answers nobody and spends nothing.
• Answer the common questions on the page itself. A chatbot that spends its allowance repeating your opening hours is expensive; a line of text on the page is free.
• Pick the multiplier deliberately. On a paid plan, a 12X model spends twelve times the credits of a 1X model for the same traffic. Check the multiplier on the AI model card before you switch.
• Test deliberately. A long afternoon of test chat is a real cost on a small allowance, and more so on a high-multiplier testing model.
• Watch for automated traffic. If usage spikes without matching conversations from real people, check where the chatbot is embedded — see Allowed domains.
• Buy a pack before a campaign, not during one. Packs never expire, so buying early costs nothing extra and avoids an outage during the busy hour.

Limits and plan notes

Credits are per account, not per chatbot. Ten chatbots share one pool. If you need to know which one is spending, use the Top Chatbots breakdown on the dashboard.

Members cannot see or buy credits. Billing is owner-only, so a member who notices the chatbot has stopped can only tell the owner. See Why Billing is missing.

Some plans are uncapped on credits. Where that applies, the meter simply never exhausts.

Support can grant credits. If you have been given a complimentary grant, the extra shows on the meter and in the sidebar as a bonus. It is not something you can arrange yourself.

Running out never deletes anything. Your chatbots, knowledge, conversations, and customers are all untouched. Answering resumes the moment credits are available again.

Common problems

Visitors say the chatbot is unavailable.

Check the credits meter before anything else. That message is what an exhausted allowance looks like from outside. See We are out of credits.

My usage is far higher than I expected.

Count your own testing first, then check each chatbot's AI model card — a model above 1X spends several credits per reply. Check whether a cached answer assumption misled you, too — cached replies are charged. Then look at Top Chatbots on the dashboard to see which chatbot is responsible.

I upgraded and my credits did not refill.

Expected. A mid-period plan change raises the allowance from your next renewal; it does not reset what you have already used. Buy a pack to get moving today.

I have pack credits left and chat is still blocked.

Reload Billing. If the balance genuinely is not zero and chats are still refused, note the approximate time and contact support. Do not include any keys or card details in the message.

Does a failed or empty answer still cost?

If the chatbot produced and delivered a reply, yes — including a reply saying it does not know. If the turn never completed, no.