# OpenAI

Connect your OpenAI account to use GPT models for the assistant, prompts, and AI rules.

{% hint style="warning" %}
Missive requires an **OpenAI API key**, not a ChatGPT subscription. These are separate services with separate billing. Get your API key at [platform.openai.com](https://platform.openai.com/).
{% endhint %}

## Setup

{% stepper %}
{% step %}
**Get your OpenAI API key**

Create an [OpenAI account](https://platform.openai.com/signup) if you don't have one.

Go to [OpenAI Platform](https://platform.openai.com/) and navigate to **Settings** > **API Keys**.

Click **Create new secret key** and copy it somewhere safe. You won't be able to see it again.

(Optional) Find your **Organization ID** in **Settings** > **General**. You'll need this if your OpenAI account belongs to multiple organizations.
{% endstep %}

{% step %}
**Add the integration in Missive**

In Missive, go to **Settings** > **Integrations** > click **Add integration** > search for **OpenAI** and click **Add to Missive**.

Enter your **API key** (required).

Under **Advanced settings** (optional):

* **Organization ID** — only needed if your OpenAI account belongs to multiple organizations
* **API Base URL** — use `https://eu.api.openai.com/v1` for EU data residency; leave blank for the default US endpoint

Click **Add**.
{% endstep %}

{% step %}
**Share with your team (optional)**

To let your team use AI features or create AI rules, share the integration:

Go to **Settings** > **Integrations** > **OpenAI** and click **Share integration** at the top.

{% hint style="warning" %}
Shared integrations mean everyone uses the same OpenAI API key. All usage bills to your OpenAI account.
{% endhint %}
{% endstep %}

{% step %}
**Restrict available models (optional)**

By default, all models are available to your team. To limit which models they can choose in the Assistant:

Go to **Settings** > **Integrations** > **OpenAI** and scroll to **Models**.

Select the models you want to make available. Choosing **All models** automatically includes any future models. Deselecting specific models prevents your team from using them.
{% endstep %}
{% endstepper %}

## Available models

### Assistant models

| Model      | Best for                                            | Speed    | Quality |
| ---------- | --------------------------------------------------- | -------- | ------- |
| GPT-5.2    | Complex reasoning, detailed analysis                | Moderate | Highest |
| GPT-5 Mini | Most daily tasks, good balance of speed and quality | Fast     | Good    |
| GPT-5 Nano | Quick tasks, lowest cost                            | Fastest  | Basic   |

For drafting customer-facing email replies, GPT-5.2 gives the best results. GPT-5 Mini is great for quick tasks like grammar fixes or summaries.

### Rule models

| Model      | Description                                           |
| ---------- | ----------------------------------------------------- |
| GPT-5 Mini | Fast, cost-effective for most tasks (recommended)     |
| GPT-5 Nano | Fastest, lowest cost option for simple classification |

## Data privacy

OpenAI does **not** use API data to train models unless you explicitly opt in. Prompts and responses are retained for up to 30 days for abuse monitoring, then deleted.

To confirm data sharing is off, go to [platform.openai.com](https://platform.openai.com/) > **Settings** > [**Data controls**](https://platform.openai.com/settings/organization/data-controls/sharing) > **Sharing** and verify all three options are **Disabled**:

* **Enable sharing of model feedback from the Platform**
* **Share evaluation and fine-tuning data with OpenAI**
* **Share inputs and outputs with OpenAI**

OpenAI also offers data residency options for organizations that need data to stay in a specific region. To use EU data residency with Missive, set the **API Base URL** to `https://eu.api.openai.com/v1` in **Advanced settings**. See our [AI FAQ](https://missiveapp.com/docs/using-ai/faq#openai) and [OpenAI's data controls documentation](https://developers.openai.com/api/docs/guides/your-data) for details.

## Cost and usage

OpenAI bills based on tokens processed (input and output). A token is roughly 3/4 of a word. Check [OpenAI's pricing page](https://openai.com/api/pricing/) for current rates.

To put it in perspective: reading a 10-email thread and drafting a reply typically uses 2,000 to 4,000 tokens. At that rate, even heavy daily use of the assistant stays well under a dollar per day with GPT-5 Mini.

{% hint style="info" %}
Set a [monthly budget limit](https://platform.openai.com/settings/organization/limits) in OpenAI to prevent unexpected charges.
{% endhint %}

### Managing costs

* **Use the best model for manual tasks.** When you trigger AI yourself (drafting a reply, asking the assistant a question), the cost per interaction is negligible. Use GPT-5.2 for the best results.
* **Use cheaper models for AI rules.** Rules run automatically on every matching message, so costs add up. GPT-5 Mini or GPT-5 Nano are better choices for classification and routing.
* **Add conditions to AI rules.** Limit when rules run (e.g., only unassigned conversations, only specific labels) to avoid processing every email.
* **Monitor usage.** Check your [OpenAI dashboard](https://platform.openai.com/usage) regularly.