Skip to main content

Automation Action: AI Assist

Peter Novosel avatar
Written by Peter Novosel
Updated this week

The AI Assist: Custom Prompt action in SmartSuite lets you integrate large language models (LLMs) like OpenAI’s GPT directly into your automated workflows. This action gives you full control over how data from your SmartSuite records is passed to AI models and how the results are structured for use in downstream automation steps.


Overview

The AI Assist: Custom Prompt action allows you to:

  • Compose a tailored prompt using dynamic field data

  • Choose the AI provider and model

  • Define expected outputs in a structured format

  • Use AI-generated results in subsequent automation steps


Adding the AI Assist Action

1. Choose a Trigger

Set up a trigger that determines when the automation runs and which record is used as the data source. Example:

  • Trigger: When a Record is Created

  • Action: Evaluate the new record with AI and extract data from it to use in subsequent automation actions.

2. Select the AI Assist: Custom Prompt Action

  1. Under Integrations, choose AI Assist.

  2. Select Custom Prompt to start configuring your AI action.


Setting Up a Connection to an AI Provider

Before using AI Assist, you must connect your SmartSuite workspace to an LLM provider such as OpenAI.

Steps:

  1. Select a Provider (e.g., OpenAI)

  2. Enter your API key

  3. Name your connection

Each Solution Manager can create their own connections. Multiple connections to the same provider are supported within a solution.

After setup, the connection becomes available to all Solution Managers in that solution. See them displayed in the Existing Connections section (screenshot below).

You may later remove connections, with a warning if automations will be affected.

For more information about connecting your AI Platform, please see this help article: AI Assist: Connecting Your Provider


Selecting a Model

Once connected, you’ll choose a model for each Custom Prompt action. Each provider displays a list of available models, such as:

  • gpt-4o, gpt-4o-mini

  • gpt-3.5-turbo

Each model includes:

  • A short description

  • Icons indicating whether the model supports:

    • Images

    • PDF Files

  • A cost indicator ($ to $$$$)

Note: You can only use one model per Custom Prompt action.


Composing the Prompt

Use a mix of static text and dynamic variables (from triggers or previous automation steps) to build your prompt.

You can reference:

  • Fields from single-record outputs (e.g., “Trigger: Product Description”)

  • Fields from multi-record outputs (e.g., “Find Orders: Order Total”)

  • Attachments and multi-value fields

Adding Fields as Reference Data

To add a field from the trigger or a previous action:

  1. Click on the blue "plus" icon in the right hand corner of the Prompt input.

  2. Select a field from the displayed selection list, OR

  3. Click the back arrow (<) in the top left corner to select another source

    1. Once clicked, you can select the trigger or a prior action as the data source

    2. After you make your selection, you can choose from a list of available data fields to insert into your prompt.

Prompt values are sent to the AI, including both the text and referenced data values.

Adding Attachments to the Prompt

To add a file to your prompt contents:

  1. Click on the blue "+" icon in the Attachments input

  2. Select the source for the attachment. You can source files from three different locations:

    1. Fields from the Trigger

    2. Fields from another Action

    3. Manually uploaded files

Prompt Examples:

  • “Translate Trigger: Product Description to Italian”

  • “Calculate the average of Action: Order Amount for the current month”

Or more complex:

Respond as a corporate compliance expert. Evaluate this policy and recommend updates based on industry best practices.

Policy: Policy Name

Purpose: Policy Purpose

Scope: Policy Scope

Statement: Policy Statement


Defining the Output

The output defines the format you expect the AI to return. You can choose from:

1. Simple Output

  • Returns raw text

  • Best for free-form responses

  • Treated as a multi-line Text Area in SmartSuite

2. Custom Output

Custom outputs can create more complex, structured data that can be used later in your automation workflow.

To create Custom output:

  • Click on the Custom tab under the Output section.

  • Define the structured data to return by clicking Add Output to build the data structure.

  • Output values can contain their own instructions that help "teach" the AI model how to format the specific value.

Configuring these Custom output values allows downstream automations to use the response in fields and logic conditions.

Custom Output Types:

  • Single Value (e.g., Title, Price)

  • List of Values (e.g., List of Emails)

  • JSON Object or Array

Each output field includes:

  • Name (required)

  • Type (Text, Number, Boolean, etc.)

  • Optional instruction to guide the AI

When using Custom Output, SmartSuite appends a system prompt to enforce JSON formatting.


Supported Field Types for Output

Custom Outputs can return any of the following:

  • Text

  • Number

  • Yes/No (Boolean)

  • Date / Time / Date & Time

  • Email

  • URL

  • Phone

  • Single Select

  • Duration

SmartSuite maps these data types automatically to ensure compatibility with downstream actions.


Defining a Custom Output

When configuring a Custom Prompt, you can define a Custom Output to structure the AI’s response in a predictable format. This makes it possible to use the returned values directly in later automation steps (e.g., updating a record or sending an email).

Custom Outputs consist of one or more output fields, each with a clearly defined type, name, and optional instructions to guide the LLM’s response formatting.

⚠️ When defining a Single Select output, you must ensure the response exactly matches one of the options configured in the target SmartSuite field. For example, if the field accepts “Open”, “In Progress”, and “Closed”, the LLM must return one of those values without variation.


Custom Output Structure Options

You can return:

  • A Single Value (e.g., translated text, calculated score)

  • A List of Values (e.g., list of customer names)

  • A JSON Object (multiple structured fields)

  • A List of JSON Objects (e.g., grouped or tabulated outputs)

Each field in your Custom Output must include the following:

Property

Description

Name

The unique label for the output field

Type

One of the supported SmartSuite field types

Instruction (optional)

Additional guidance for the AI to produce the correct format

Example: JSON Object Output

{ "Response Format": { "Type": "Object", "Name": "Response", "Properties": [ { "Name": "Priority", "Type": "Single Select", "Instruction": "Set to one of the following: High, Medium, Low" }, { "Name": "Due Date", "Type": "Date", "Instruction": "Return in YYYY-MM-DD format" } ] } }

Tips for Formatting Output Correctly

  • Dates & Times: Use ISO 8601 format to ensure SmartSuite parses them correctly.

  • Single Select Fields: Match values exactly, including case and spelling.

  • Phone Numbers: Consider using a normalized format like +1 (123) 456-7890.

  • Lists of Values: Return each item as an array entry, not a comma-separated string (unless instructed).

Example: List Output

{ "Response Format": { "Type": "List", "Name": "Top Customers", "List Item": "String", "Instruction": "Return the top 5 customer names as an array" } }

Adding Output Instructions

While not required, adding clear Instructions improves accuracy and helps guide the AI model in generating valid data types.

Examples:

  • "Return in YYYY-MM-DD format."

  • "Choose from one of the following: Yes, No."

  • "Generate a list of values using the +1 (XXX) XXX-XXXX phone format."

These instructions are not included in the output but are used to shape the AI’s response.


Validating and Using Custom Outputs

Once defined, Custom Outputs are:

  • Validated during configuration and test runs

  • Available in later automation steps via variable pills

  • Mapped into SmartSuite fields, provided the formats align

If a Custom Output is not compatible with the selected SmartSuite field, automation will show a configuration error.

Including precise output definitions ensures reliable AI behavior and smooth integration into SmartSuite workflows. Always test with realistic data before enabling your automation live.


Using the Output in Automations

Once the Custom Prompt returns data, you can use that output in any of the following actions:

  • Create / Update / Delete Record

  • Send Email

  • Find Records

  • Conditional Branching

  • Looping

Each returned value becomes available for field mapping or decision-making logic.


Working with Attachments

If the source data includes a file (such as a PDF resume), SmartSuite will attempt to send that file to the LLM—provided the model supports file input. Ensure your prompt explicitly references the file.


Example Use Cases

Trigger

Prompt

Output Format

New resume submitted

Summarize resume with candidate name, experience, and skills

JSON object with structured fields

Order approved

Convert “Order Total USD” to local currency

JSON number

Quote dataset retrieved

Group quotes by contact

List of objects with contact names and quote IDs

Email received

Categorize email topic

Single select string output (e.g., Sales, Support)


Best Practices

  • Use Custom Outputs when results are needed for additional automation steps.

  • Include clear instructions in the prompt to improve accuracy.

  • Test your prompt and model selection before enabling the automation.

  • For sensitive data, ensure you are using secure API keys and understand the provider’s data policies.


Troubleshooting

  • Prompt is invalid: Make sure all required fields are added and that the prompt is not empty.

  • Output not usable: Check that the Custom Output format aligns with the fields you want to use.

  • LLM model limitations: Some models may not support attachments or large data sets.


FAQs

Can I use multiple LLMs in one automation?
No. Each Custom Prompt action can use only one model at a time.

Can I switch models later?
Yes, but SmartSuite will validate compatibility with your prompt and output settings.

Can I use multiple Custom Prompts in a single automation?
Yes. Each step can have its own prompt, model, and output structure.

Did this answer your question?