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

# Included AI

> Let assistants use the platform's own AI account, with a fair-use allowance per assistant set by each plan, so agencies don't need their own AI keys.

With included AI, the platform owner connects one AI account, and an agency whose plan includes AI can set its assistants to **Included with the plan**. Those assistants need no API key and no model choice. Each plan sets a monthly fair-use allowance per assistant, and the agency's assistants share it.

Agencies whose plan doesn't include AI work as before: each assistant uses its own key.

## Set it up (platform owner)

Open the **Platform** page and find **Included AI**.

<Steps>
  <Step title="Choose the service and model">
    Pick the AI service and type the exact model name. A small, fast model is usually best for customer chat. With **OpenAI-compatible**, also fill in the base URL.
  </Step>

  <Step title="Paste the API key and save">
    The key is stored encrypted and is never shown again. Usage is billed to this account.
  </Step>

  <Step title="Click Test connection">
    You should see "Connected. The model replied." If not, check the key and the model name.
  </Step>
</Steps>

| Setting | What it does |
| - | - |
| Offer included AI | Off: assistants set to it stop answering with AI until it's back on, and new assistants need their own key |
| Input price, output price | USD per million tokens, used to count each agency's use. Leave empty for the model's list price |
| Pause at | The share of the allowance at which included AI pauses until the 1st. 150% unless you change it |

Then give your plans an allowance: in each plan, fill in **Included AI per assistant (USD a month)**. Leave it empty for a plan without AI. An agency on no plan can get an allowance of its own in its edit window.

<Info>
  The allowance is counted at list prices, the same way each assistant's AI cost is shown. A typical answer costs a fraction of a cent on a small model, so an allowance of a few dollars per assistant covers thousands of answers. Check the **Use this month** table after the first weeks and adjust.
</Info>

## How fair use works

* **Shared by the agency.** The agency's allowance is the per-assistant amount times its assistants (demos aside, at least one). A busy client can use what a quiet one doesn't.
* **Alerts.** At 80% and at 100% the agency's admins get an email. At 100% you get an alert too.
* **Over 100%,** assistants keep answering. At the **Pause at** level, included AI pauses until the 1st: the assistants answer with their fallback message, and rules, FAQs and question flows keep working. The hourly [channel check](/running-the-agency/channel-checks) shows it as a problem.
* **Lifting a pause.** A bigger allowance, a bigger plan, or the agency's own key on an assistant's **AI brain** page lifts it straight away.
* **Every month,** the count starts again on the 1st.

The **Use this month** table on the Platform page shows each agency's allowance, what it spent and the share used.

## What agencies see

When their plan includes AI:

* New assistants start on **Included with the plan**. Existing ones can be switched on the [AI brain](/build-the-assistant/ai-brain) page.
* The AI brain page and **Settings → Plan** show how much of this month's allowance is used, as a percentage. Agencies never see dollar amounts or your key.
* Voice notes and photos are read by the included AI when its model can do that.

## Privacy

Each assistant's privacy page names the company that really processes the messages: the service you chose for included AI, not "the platform". If that service processes messages outside the EU, the page says so, as it does for an assistant's own key.

## With an AI agent

Platform owners can ask Claude or another AI agent connected to the platform for **included\_ai** (the settings and each agency's use), and set allowances with **save\_plan**, **create\_agency** and **update\_agency** (`ai_allowance_usd`). Agency admins see their share used in **my\_plan**. The key itself is entered on the Platform page only.


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