What Uses Credits
Billing Overview explains the credit model. This page is the detailed answer to the question people actually ask: “why did this conversation cost more than I expected?”
Short version: your AI Quality tier sets the price of one AI action, and a single incoming message can produce more than one AI action. That is normal, it is not a bug, and every action is written to your usage history as its own line so you can see exactly what happened.
What Counts as One “AI Action”
A credit is charged when the AI thinks, not when a message is sent or received. One AI action is one call to the AI.
The price of one action depends on the AI Quality tier set on the Agent:
| AI Quality tier | Credits per action | With your own Anthropic key connected |
|---|---|---|
| Pro | 1 | 0 |
| Economy (deprecated) | 0.5 | 0 |
| Max | 0.25 | still 0.25 |
| Mini | 0.15 | still 0.15 |
Max and Mini run on our own model rather than your key, which is why they still cost credits when you have brought your own key. See Billing Overview.
Why One Message Can Cost More Than One Action
Writing the reply is only one of the jobs the AI does on a conversation. Depending on what you have switched on, the same incoming message can also trigger:
| What happens | Shows in your usage history as | When it is charged |
|---|---|---|
| The AI writes a reply | AI Message | Once per reply cycle, however many bubbles the reply is split into |
| The AI uses a tool (books an appointment, checks your calendar, alerts a human, schedules a follow-up, searches your website or the web, runs one of your custom functions) | AI Tool Use: tool name | Once per tool call, on top of the reply |
| The AI decides whether to stay quiet on a message that does not need an answer | AI Chat Ignore Evaluation | Only when it actually decides to stay quiet |
| The AI decides the conversation is finished | AI Chat Conclude Evaluation | Only when it actually concludes the chat |
| A new message arrives while the AI is mid-reply and it decides whether to restart | AI Interruption Handling | Only when a message interrupts a reply |
| The AI applies a tag to the contact | Contact Tag / Chat Tag Evaluation | Once per tag applied |
| The AI writes a running summary of a long conversation | Chat Summary Generation | On long conversations only |
| The AI decides whether a follow-up is needed | Follow-Up Evaluation | When follow-ups are enabled on the Agent |
| A follow-up message is written and sent | Follow-up Cycle n | Once per follow-up sent |
| The AI pulls a name, email or phone number out of the conversation | AI First Name Extraction, AI Email Extraction, and similar | When the detail appears in the conversation |
| The spam filter checks a first message from a new contact | (no line) | Never, spam filtering is always free |
Each of those lines is priced at the same tier rate as the reply itself. So a conversation where the AI answers, books a meeting, tags the contact and decides to conclude is four actions, not one.
This is the single most common billing surprise, and the fix is usually one of these:
- Move the Agent to the Max or Mini tier. That is the biggest single lever: the same conversation costs a quarter of the Pro price on Max, and less again on Mini.
- Turn off the automations you are not using on that Agent, such as auto-tagging and follow-ups. Each one you switch off removes its line from every conversation.
- Read the Model column in your usage history (see below) before changing anything else.
How to Read Your Usage History
Settings → Billing, then scroll to Usage history. Every credit-consuming action is one row, with the date, the reason, the Agent, the contact, the Model column, and the credits charged.
The Model column is the one to check when a charge looks wrong. It shows the rate the row was actually billed at, not the tier you picked in the editor. If a row reads a different tier than the Agent you set, that is the answer to “why was I charged 1 instead of 0.25”, and it is worth sending to support with the row’s date and contact so we can look at it.
To see what a single conversation cost, sort or scan by the contact name in that table: every row for that contact, added up, is that conversation’s total.
Monthly Channel Costs
Some channels carry a fixed monthly cost because we pay for infrastructure on your behalf. Most do not.
| Channel | Monthly cost | Per-message cost |
|---|---|---|
| A phone number you rent here | At least 50 credits a month; premium numbers cost more. The number’s card on Settings → Channels shows its exact price | WhatsApp template fees passed through at cost |
| WhatsApp Web | 50 credits a month per connected number | None |
| WhatsApp connected through our Meta app | 50 credits a month per connection | Meta bills its own message fees to the WhatsApp account directly |
| Instagram DMs | None | None |
| Facebook Messenger | None | None |
| Website chat widget | None | None |
| Telegram | None | None |
| LINE | None | None |
| Viber | None | None |
| None | None | |
| SMS | None from us | Billed by Twilio on your own account, see SMS Provider |
So the answer to “do Instagram, Messenger and website chat have a monthly channel fee like WhatsApp Web does?” is no. Only rented numbers and the two WhatsApp connection types carry one.
Other Things That Use Credits
- Daily Summaries: one AI action per generated summary, plus one more for each task the summary creates from your conversations. A day with no conversations costs nothing. See Daily Summaries.
- Broadcasts: sending is free. There is no per-contact charge for launching a broadcast. You are charged only for the AI replies to people who answer, and for WhatsApp Business API template delivery fees where they apply. See Billing Overview.
- Comment-to-DM: the public reply and first DM cost one AI action each in AI mode, and 0 credits in Exact Text mode. See Comment-to-DM.
- Automations: 0.25 credits when a run starts, plus 0.25 for each AI step inside it. These are charged even with your own AI key connected.
- Knowledge base generation: importing a website or a document runs the AI over each chunk of content, and each chunk is an AI action. A large site is the one setup step that can use a noticeable number of credits in one go, so start with the pages that matter rather than a whole domain.
- Understanding incoming media: voice notes, images and documents a contact sends are read as part of the reply that handles them, at no extra charge. See Billing Overview.
A Worked Example
A contact messages your WhatsApp Web number three times in a row, asks a question, and books a call. Your Agent is on the Max tier with auto-tagging on.
| What happened | Credits |
|---|---|
| The three messages are batched into one reply cycle | 0.25 |
| The AI checks your calendar | 0.25 |
| The AI books the appointment | 0.25 |
| The AI tags the contact “booked” | 0.25 |
| The AI concludes the chat | 0.25 |
| Total for the conversation | 1.25 |
On the Pro tier the same conversation is 5 credits. That difference is why we recommend Max as the default.
Next Steps
- Billing Overview: the credit model, top-ups, auto-recharge and invoices.
- Plans & Pricing: what each plan includes.
- AI Model & BYOK: connecting your own AI key.