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

# AI Fields

> AI fields are custom fields on contacts and other records whose values are automatically generated by AI based on a prompt you define and the record's existing data.

AI fields let you define a custom field on any contact or record and have Orbis automatically fill in its value using AI. Instead of manually writing a summary, classification, or derived insight for each contact, you write a prompt once and AI generates the value for every record.

## What AI fields are

An AI field is a custom field of type **AI Generate**. When you create one, you define:

* A **prompt** describing what you want the field to contain
* **Source attributes** — the existing fields Orbis should read to generate the value

When the field is generated, the AI reads the contact's current data — including profile information, notes, and interaction history — and fills in the field value based on your prompt.

**Example:** You could create an AI field called "Relationship summary" with the prompt "Write a one-sentence summary of my relationship with this person based on our interactions and notes." Orbis generates a personalized summary for each contact automatically.

## Creating an AI field

<Steps>
  <Step title="Open a contact or record">
    Navigate to a contact's detail page, or open a record in a custom object.
  </Step>

  <Step title="Go to custom fields">
    Scroll to the **Custom Fields** section of the record, or open the fields panel.
  </Step>

  <Step title="Add a new field">
    Click **Add field** or the **+** button to open the field creation dialog.
  </Step>

  <Step title="Choose AI Generate as the field type">
    In the field type picker, select **AI Generate** (in the Computed category).
  </Step>

  <Step title="Write your prompt">
    Enter a prompt describing the value you want generated. Be specific about format and length.

    **Example prompts:**

    * "Write a one-sentence description of what this person does professionally."
    * "Based on our interactions and notes, categorize this contact as Warm, Neutral, or Cold."
    * "List the top 3 topics we've discussed, as a comma-separated list."
    * "What industry does this person work in? Return a single word or short phrase."
  </Step>

  <Step title="Select source attributes">
    Choose which fields from the record Orbis should use as context when generating the value. For contacts, this might include name, company, title, notes, and interaction history.
  </Step>

  <Step title="Save and generate">
    Save the field. To generate the value for this record, click the **Generate** button that appears next to the field.
  </Step>
</Steps>

<Tip>
  The more specific your prompt, the more consistent the results. If you want a short value, say so — "Return a single word" or "Write no more than one sentence."
</Tip>

## Supported field types

AI fields always output a **text value**. This makes them suitable for:

* Short summaries and descriptions
* Classifications and categories (e.g., "Warm / Neutral / Cold")
* Derived labels and tags
* Free-form insights

For numeric outputs, structured lists, or checkboxes, consider using a formula field instead.

## How AI generates the value

When you trigger generation, Orbis:

1. Reads the source attributes you selected — for example, the contact's name, company, title, notes, and recent interactions
2. Passes that data to the AI along with your prompt
3. Returns a concise value based on the instructions in the prompt

The AI does not browse the web or access data outside Orbis. It only uses the record's existing data within your workspace.

<Note>
  AI field generation uses Orbis's built-in AI model. For fields where you need higher accuracy, an upgraded model option is available in the field's advanced settings.
</Note>

## Rate limits on generation

AI field generation is subject to usage limits:

* **Single record generation** — available at any time, subject to your plan's AI usage quota
* **Batch generation** — when generating across many records simultaneously, requests are processed in batches of 5 records in parallel

If you hit your usage limit, Orbis will show a notification. Limits reset daily.

## When AI fields refresh

AI fields do **not** update automatically when the underlying record changes. You control when generation runs:

| Trigger           | How it works                                                                                             |
| ----------------- | -------------------------------------------------------------------------------------------------------- |
| **Manual**        | Click the **Generate** button on the field to regenerate the value at any time                           |
| **Source change** | When configured, the value regenerates automatically when any of the field's source attributes change    |
| **Schedule**      | When configured, the value regenerates on a recurring basis (available in the field's advanced settings) |

The default is **manual** — the AI generates the value when you explicitly request it.

<Info>
  Generated values are stored like any other field value. You can edit them manually after generation if the AI output needs adjusting.
</Info>

## Batch generation

If you want to generate an AI field's value for all contacts (or a filtered subset), you can trigger batch generation from the field's settings:

<Steps>
  <Step title="Open the field settings">
    Go to **Settings → Custom Fields**, or open the field configuration from the record.
  </Step>

  <Step title="Select records">
    Choose whether to generate for all records or apply a filter first.
  </Step>

  <Step title="Run batch generation">
    Click **Generate for all**. Orbis processes records in parallel and updates the field value as each one completes.
  </Step>
</Steps>

<Warning>
  Batch generation across a large number of records consumes AI usage quota proportional to the number of records. For large datasets, consider filtering to a subset first.
</Warning>
