> ## Documentation Index
> Fetch the complete documentation index at: https://docs.callivox.com/llms.txt
> Use this file to discover all available pages before exploring further.
# How credits are spent
> The conversation window, effort tiers and rates that decide what your AI work costs, with worked examples you can check.
Credits are what AI work costs. This page explains exactly when a credit is taken, how much, and
what is free — so you can predict a month's spend before it happens.
The short version: **credits burn only when an AI agent answers.** One conversation costs
**1, 3 or 10 credits per 24 hours**, depending on the effort level of the agent that answered — no
matter how many messages that conversation contains. Voice calls and knowledge-base uploads are
charged separately on top.
## What a credit is
A credit is a whole unit of AI usage. There are no fractions of a credit. On current plans one
credit is worth **\$0.14**, which is why the Credits card shows an approximate money value next to
your balance.
The card tells you three things:
* **Your balance** — spendable credits, with the approximate value beside it.
* **Held for in-flight** — credits reserved for work happening right now. This only appears when
it is above zero. See [Holds](#holds-credits-reserved-while-work-is-in-flight) below.
* **Monthly allowance** — how many credits your plan grants, and when they next arrive.
## A conversation is the billing unit
Callivox does not charge per message. It charges per **conversation**, per **24-hour window**.
A conversation is one contact on one channel — for example, a single WhatsApp thread with one
customer. The first time an AI agent answers in that conversation, Callivox:
1. Charges the conversation fee for that agent's effort level, and
2. Opens a **24-hour window** on that conversation.
For the rest of that window, every further AI answer in the same conversation costs **nothing**.
Ten messages or two hundred, over ten minutes or twenty hours — same fee.
When the window expires, nothing happens by itself. The next time an AI agent answers in that
conversation, a new window opens and the fee is charged again.
### Worked example: one thread over two days
A customer messages your WhatsApp channel. A **Medium** effort agent is on that channel.
| When | What happens | Credits |
| ------------------ | --------------------------------------------------------------------------------------- | -------------------------------------------------- |
| Mon 3 March, 09:12 | Customer asks about a delivery. The AI answers. | **3** — window opens, runs until Tue 4 March 09:12 |
| Mon 3 March, 09:14 | Customer replies. The AI answers. | 0 |
| Mon 3 March, 11:40 | Customer sends a photo of the parcel. The AI answers. | 0 |
| Mon 3 March, 16:55 | A human agent takes over and replies twice. | 0 |
| Tue 4 March, 08:50 | Customer follows up. The AI answers — still inside the window. | 0 |
| Tue 4 March, 10:05 | Customer writes again. The window expired at 09:12, so the AI's answer opens a new one. | **3** — new window until Wed 5 March 10:05 |
Total for the whole two-day thread: **6 credits**.
### What is free
None of these open a window or cost a credit:
* **Inbound messages** from customers, on any channel
* **Replies typed by your human agents**
* **Scripted flow steps** — the deterministic parts of a flow agent
* **Sending and receiving messages** in general — message delivery itself carries no credit cost
That means a busy inbox staffed by people costs nothing in credits. Only AI answers do.
## Effort tiers
Every AI agent has an effort level — **Low**, **Medium** or **High** — which you set on the agent.
Higher effort means the platform uses more capable processing for that agent, and costs more per
conversation.
| Effort | Credits per conversation, per 24-hour window |
| ---------- | -------------------------------------------- |
| **Low** | 1 |
| **Medium** | 3 |
| **High** | 10 |
The fee is decided by the agent that answered, not by the channel or the customer. If two
different agents answer in the same conversation, see below.
### When a higher-effort agent joins mid-window
If a conversation already has an open window and an agent with a **higher** effort level answers,
the window is upgraded to that level and you are charged only the **difference**.
A worked example, on one conversation:
| When | What happens | Credits |
| ----- | --------------------------------------------------------------------- | -------------------------------------------------------- |
| 14:00 | A **Low** agent answers. | **1** — window opens at Low |
| 14:20 | The same Low agent answers again. | 0 |
| 17:40 | The conversation is handed to a **High** effort agent, which answers. | **9** — the difference between 10 and the 1 already paid |
| 18:05 | The High agent answers again. | 0 |
| 21:00 | A **Low** agent answers — lower than the window's level. | 0 |
Total: **10 credits** — the same as if the High agent had answered first. A window is never
charged twice for the same level, and never charged down.
## The full rate card
These are the rates on current plans.
| What | Rate | How it is counted |
| --------------------------------------------------------------------- | ---------- | --------------------------------------------- |
| Conversation — **Low** effort | 1 credit | Per conversation, per 24-hour window |
| Conversation — **Medium** effort | 3 credits | Per conversation, per 24-hour window |
| Conversation — **High** effort | 10 credits | Per conversation, per 24-hour window |
| Voice call connect | 2 credits | Once per call |
| Voice minute — **Low** effort | 2 credits | Per minute, rounded up |
| Voice minute — **Medium** effort | 4 credits | Per minute, rounded up |
| Voice minute — **High** effort | 8 credits | Per minute, rounded up |
| Knowledge-base ingestion | 1 credit | Per 1,000 tokens of document text, rounded up |
| WhatsApp / Instagram / Telegram / TikTok / email / web-widget message | 0 | Messages are not charged |
| Inbound message | 0 | Not charged |
| Outbound message, human reply | 0 | Not charged |
| AI image | 1 credit | Included in the conversation fee |
| AI document | 3 credits | Included in the conversation fee |
| AI tool call | 1 credit | Included in the conversation fee |
The last three lines are listed on the rate card because Callivox can price them individually on
custom plans. On the standard plans they are **covered by the conversation fee** — if a customer
sends your AI agent an image or a PDF inside a conversation, or the agent looks something up in
one of your connected systems, you pay the conversation fee and nothing more.
**Voice calls and knowledge-base ingestion are always charged on top** of the conversation fee.
They are not covered by the 24-hour window.
## Voice calls
A voice call is charged in two parts:
* **A connect fee of 2 credits**, once, when the call connects.
* **A per-minute rate**, set by the voice agent's effort level: 2, 4 or 8 credits a minute.
Minutes are rounded **up**: a call lasting 2 minutes 10 seconds is charged as 3 minutes.
If a customer both calls you and messages you, those are two separate things — the call is charged
as above, and the message thread is charged as its own conversation.
## Knowledge-base documents
When you upload a document to a knowledge base, Callivox reads it once and indexes it so your
agents can answer from it. That costs **1 credit per 1,000 tokens** of text in the document,
rounded up. A token is a short piece of text, a bit under a word on average.
You pay this once per document, when it is ingested — not every time an agent uses it. Answering
from your knowledge base during a conversation costs nothing beyond the conversation fee.
## Holds: credits reserved while work is in flight
Callivox reserves credits before it starts chargeable work, then settles them when the work
finishes. This is what **Held for in-flight** on the Credits card means.
Before an AI agent starts working, the credits it will need are set aside. Your balance is
unchanged, but those credits are no longer spendable by anything else. If you cannot afford
them, the work does not start — which means you are never charged for an answer you did not
get.
When the work completes, the reserved credits are taken from the balance and a line is written
to your ledger.
If the work does not happen, the hold is dropped and nothing is charged.
Holds are short-lived. If one is ever left behind — for example after an interruption — Callivox
clears it automatically, so held credits never leak away from your balance.
Your true spendable amount is **balance minus held**. If a top-up looks like it has not landed,
check whether credits are being held for work in progress.
## Worked examples
### A support team on WhatsApp
A team handles **400 WhatsApp conversations a month**, all answered by a single **Medium** effort
agent. Most threads run over an hour or two; a handful spill into a second day.
```
400 conversations × 3 credits = 1,200 credits
```
That fits inside the **Growth** plan's 5,000 monthly credits with 3,800 left over — which roll
over into next month. It also fits inside **Starter**'s 1,000 credits only if volume drops, so
Growth is the safer choice at that level.
If 30 of those 400 threads carry on past their 24-hour window and get a second AI answer the next
day, add 30 more windows:
```
430 windows × 3 credits = 1,290 credits
```
### A mixed workspace
An online retailer runs three agents and refreshes its knowledge base once a month.
| Usage | Maths | Credits |
| ----------------------------------------------------------------- | -------------- | --------- |
| 600 order-status conversations on a **Low** agent | 600 × 1 | 600 |
| 120 pre-sales conversations on a **Medium** agent | 120 × 3 | 360 |
| 40 complaint conversations on a **High** agent | 40 × 10 | 400 |
| Monthly knowledge-base refresh, 62,000 tokens of policy documents | 62,000 ÷ 1,000 | 62 |
| Customers sending photos and PDFs into those conversations | included | 0 |
| Human agents replying to 350 of the same conversations | included | 0 |
| **Total** | | **1,422** |
Comfortably inside **Growth** (5,000 credits). Note how much of the volume is free: only the AI's
first answer in each 24-hour window is charged.
### A voice team
An appointments team runs one **Medium** effort voice agent, handling **120 calls a month**,
averaging 3 minutes 20 seconds each — charged as 4 minutes.
| Part | Maths | Credits |
| ----------- | --------------------------------- | --------- |
| Connect fee | 120 × 2 | 240 |
| Minutes | 120 calls × 4 minutes × 4 credits | 1,920 |
| **Total** | | **2,160** |
Voice is the most credit-hungry thing Callivox does. If those same calls ran on a **Low** effort
voice agent, the minutes would cost 120 × 4 × 2 = 960 credits, bringing the total to 1,200.
## Spending less
* **Match effort to the job.** Order lookups and FAQs rarely need High effort. Moving a
high-volume agent from High to Medium cuts its conversation fee by 70%.
* **Let humans and flows do the routine parts.** Scripted flow steps and human replies are free,
and a flow agent that only calls on AI when the script runs out costs less than a free-form one.
* **Do not fear long threads.** Resolving something fully in one conversation costs the same as
answering once. There is no saving in cutting a conversation short.
* **Ingest knowledge bases deliberately.** Re-uploading the same document re-ingests it and is
charged again, so keep refreshes to what actually changed.
## Where to see what you spent
The **Recent activity** table on **Settings → Billing** is your credit ledger. Every grant,
purchase and charge appears there, newest first.
| Column | What it shows |
| ----------- | ------------------------------------------------------------------------------------------------------------------ |
| **Date** | When the credits moved |
| **Type** | `trial grant`, `monthly refill`, `topup`, `debit` or `adjustment` |
| **Credits** | Green `+N` when credits arrive, red `-N` when they are spent |
| **Reason** | What was charged — for example a conversation at a given effort level, a voice call, or a knowledge-base ingestion |
Credits arriving show as **monthly refill** (your plan's grant), **topup** (a purchase) or
**trial grant**. Everything spent shows as a **debit**.
## Next steps
Buy more credits, or keep the balance topped up automatically.
What you are warned about, and exactly what stops at zero.
Match your monthly credit need to the right plan.
Where you set an agent's effort level.