> ## Documentation Index
> Fetch the complete documentation index at: https://docs.qall.io/llms.txt
> Use this file to discover all available pages before exploring further.

# What a call costs

> Where the money goes, and the levers that move it.

A call is billed on what it used: the language model, the voice, the
transcription, and a platform charge for the minutes.

You can see it two ways — an estimate before you commit, and the real figure
afterwards.

## Before: the estimate

The assistant editor shows an estimated **cost per minute** as you pick models,
split by language model, voice and transcription. Change a model and the figure
moves.

Use it to compare before you build, not after you've run a thousand calls.

## After: the real figure

Every [call](/calls) carries its actual cost, broken down the same way. The
**Analytics** tab charts it — cost grouped by assistant tells you which line is
expensive, and cost per minute over time tells you whether a change helped.

## The levers, roughly in order

### 1. The language model

Usually the biggest line, and the easiest to overspend on. A smaller model often
makes a *better* phone assistant — on a call, a fast adequate answer beats a slow
excellent one.

Try the cheapest model that holds the conversation, then move up only where it
fails.

### 2. Your own provider keys

Add your own keys under [Integrations](/integrations) and Qall doesn't bill you
for that provider's usage — you pay them directly. If you already have volume
pricing with OpenAI or ElevenLabs, this is usually the single biggest saving
available.

You can do it per provider. Your own ElevenLabs key with Qall's language model is
a normal setup.

### 3. How long calls run

Minutes are charged, so anything that ends a call sooner saves money across the
board.

* **Keep replies short.** "One or two sentences" in the prompt. Long answers cost
  model output, voice characters and minutes at once.
* **Set an idle timeout**, so an abandoned call doesn't hold the line.
* **Let the assistant end the call** when the conversation is finished — give it
  an end-call tool.

### 4. Knowledge in the prompt

A document set to **always in the prompt** is re-sent on every turn of every
call. Three long documents can cost more than the conversation. Use **search**
unless the document is short and always relevant. See [Knowledge](/knowledge).

### 5. Outcomes

Each [outcome](/outcomes) assigned to an assistant adds a little work after every
one of its calls. Record what you'll act on. Five well-chosen outcomes are worth
more than fifteen you never chart.

## What doesn't move the needle

**A shorter system prompt.** It's small next to the conversation itself.

**Switching voice provider to save money.** The voice is rarely the biggest line,
and a voice your callers dislike costs you more in hang-ups than you save.

## Keeping the lights on

Qall runs on prepaid credit. Turn on **auto-recharge** under
[Organisation](/organisation) so a busy day doesn't run you to zero.

Watch cost per minute rather than total spend — total goes up when you're busy,
which is good news. Cost per minute going up is the thing to look at.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.