AI Service Layer API

For pricing information, please refer to Domo’s online consumption terms and the credit rate card in your Domo instance (Admin > Company Settings > Credit Utilization > Rate Card) for more information.

Domo’s AI Service Layer enables developers to build AI capabilities into their Domo Apps. In particular, the AI Service Layer currently supports two services from within Apps:

1. Text Generation
2. Text-to-SQL
3. Image-to-Text

Domo allows you to configure which models power each of the services. For more on how that works, see this video.

You can also see example usage of the the Text Generation and the Text-to-SQL in the AI Domo Bricks currently available to download for free in the Domo AppStore.

• AI ChatGPT Brick
• ChatGPT Dataset Description Brick
• ChatGPT Text-To-SQL Query Brick

Text Generation

---

Generates a text response from a text prompt.

Code Example

The body variable in this post request is an example of a sample request body.

const prompt = 'Tell me a joke about data.';
const body = {
  input: prompt,
};

domo
  .post(`/domo/ai/v1/text/generation`, body)
  .then((response) => console.log(response));

Arguments

Property Name	Type	Required	Description
input	String	Required	The prompt you are sending to the model
promptTemplate	Object	Optional	An override for the prompt template used in the service. It has one property template, which expects a string.
parameters	Object	Optional	Used with the promptTemplate for additional customization. You can pass any key, value pair of strings. See examples below.
model	String	Optional	The id of the model you’d like to use if you don’t want to use the default model.

HTTP Request

POST /domo/ai/v1/text/generation HTTP/1.1
Content-Type: application/json
Accept: application/json

Request Body

The only required field in the request body is the input string, but you can provide additional properties in the request body to futher customize the text generation service. If you choose to use the promptTemplate property to override the default prompt template, you’ll need to use ${input} into the template string to pass through the input of your prompt.

{
  "input": "Recap the 2021 superbowl",
  "promptTemplate": {
    "template": "${input}. You are a helpful assistant that gives answers in ${max_words} words or less"
  },
  "parameters": {
    "max_words": "30"
  },
  "model": "8dc5737d-0bc8-425b-ad0d-5d6ec1a99e72"
}

HTTP Response

Returns the prompt used and the responses from the model. If the a model was specified in the request, then modelId will return the model and isCustomerModel will be true.

HTTP/1.1 200 OK
Content-Type: application/json;charset=UTF-8

{
    "prompt": "Recap the 2021 superbowl",
    "choices": [
      {
        "output": "The 2021 Super Bowl, also known as Super Bowl LV, took place on February 7, 2021 and James Stadium ..."
      }
    ],
    "modelId": "8dc5737d-0bc8-425b-ad0d-5d6ec1a99e72",
    "isCustomerModel": true

}

Text-to-SQL

---

Generates a SQL query based on a prompt and a DataSet’s schema.

Code Example

The body variable in this post request is an example of a sample request body.

const exampleDataSourceSchema = {
  dataSourceName: 'Sales',
  description: 'Sales Data',
  columns: [
    {
      name: 'Date',
      type: 'date',
    },
    {
      name: 'Sales',
      type: 'number',
    },
  ],
};
const body = {
  input: 'Show me the sales for the last 3 months.',
  dataSourceSchemas: [exampleDataSourceSchema],
};

domo
  .post(`/domo/ai/v1/text/sql`, body)
  .then((response) => console.log(response));

Arguments

Property Name	Type	Required	Description
input	String	Required	The prompt you are sending to the model
promptTemplate	Object	Optional	An override for the prompt template used in the service. It has one property template, which expects a string.
parameters	Object	Optional	Used with the promptTemplate for additional customization. You can pass any key, value pair of strings. See examples below.
model	String	Optional	The id of the model you’d like to use if you don’t want to use the default model.
dataSourceSchemas	Array of Objects	Optional	The schemas of datasets that they service should take into account when generating the SQL query from the input prompt.

HTTP Request

POST /domo/ai/v1/text/sql HTTP/1.1
Content-Type: application/json
Accept: application/json

Request Body

The only required field in the request body is the input string, but you can provide additional properties in the request body to futher customize the text generation service. If you choose to use the promptTemplate property to override the default prompt template, you’ll need to use ${input} into the template string to pass through the input of your prompt.

{
  "input": "Create a sql query to show me total sales.",
  "promptTemplate": {
    "template": "${input}. Show me the ${measure} for the last ${timeframe}"
  },
  "parameters": {
    "measure": "sales",
    "timeframe": "3 months"
  },
  "model": "8dc5737d-0bc8-425b-ad0d-5d6ec1a99e72",
  "dataSourceSchemas": [
    {
      "dataSourceName": "Sales",
      "description": "Sales Data",
      "columns": [
        {
          "name": "Date",
          "type": "date"
        },
        {
          "name": "Sales",
          "type": "number"
        }
      ]
    }
  ]
}

HTTP Response

Returns the prompt used and the response options from the model. If the a model was specified in the request, then modelId will return the model and isCustomerModel will be true.

HTTP/1.1 200 OK
Content-Type: application/json;charset=UTF-8

{
    "prompt": "You are a helpful assistant that generates SQL queries. Table `Voter Registration`, columns=[State:string, County:string, Party:string, Registered Voters: number, Total Votes:number]. Find the total votes by party in each county in the state of Utah. Use column aliases only for functions. Do not elaborate.",
    "choices": [
      {
        "output": "SELECT County, Party, SUM(`Total Votes`) AS TotalVotes FROM `Voter Registration` WHERE State = 'Utah' GROUP BY County, Party"
      }
    ],
    "modelId": "8dc5737d-0bc8-425b-ad0d-5d6ec1a99e72",
    "isCustomerModel": true
}

Image-to-Text

---

Generates a text response from an image.

Code Example

The body variable in this post request is an example of a sample request body.

export async function imageToText(base64) {
  // Example: OCR request body for AI API
  // Remove data URL prefix if present
  let pureBase64 = base64;
  if (pureBase64.startsWith('data:')) {
    pureBase64 = pureBase64.substring(pureBase64.indexOf(',') + 1);
  }
  const body = {
    input: 'return the text in the image, ensuring new lines are preserved',
    image: {
      mediaType: 'image/png',
      type: 'base64',
      data: pureBase64,
    },
    model: 'domo.domo_ai.domogpt-chat-medium-v1.1:anthropic',
    promptTemplate: {
      template: '${input}',
    },
    system: `You are an AI assistant tasked with performing Optical Character Recognition (OCR) on an image. Your goal is to accurately identify and transcribe any text present in the image by analyzing the image carefully and extract all visible text.\n\nFollow these steps to perform OCR on the image\n\n1. Examine the entire image thoroughly, paying attention to all areas where text might be present.\n2. Identify any visible text, including numbers, letters, symbols, and punctuation marks.\n3. Transcribe the text exactly as it appears in the image, maintaining the original spelling, capitalization, and punctuation.\n4. If the text is arranged in multiple lines or paragraphs, preserve this structure in your transcription.\n\nRemember, your task is to transcribe the text as accurately as possible without interpreting or summarizing its content. Focus solely on the text visible in the image.\n\nBefore responding think about your response step by step. Provide your final transcription within <answer> tags. For example\n<example>\n<answer>\nHello World\n</answer>\n</example>\n`,
  };
  const response = await fetch('/domo/ai/v1/image/text', {
    method: 'POST',
    headers: {
      'Content-Type': 'application/json',
    },
    body: JSON.stringify(body),
  });
  if (!response.ok) {
    throw new Error('Image-to-text request failed');
  }
  return await response.json();
}

Arguments

Property Name	Type	Required	Description
input	String	Required	The prompt you are sending to the model (e.g., ‘return the text in the image, ensuring new lines are preserved’)
image	Object	Required	The image object containing mediaType, type, and data (base64-encoded image).
model	String	Required	The id of the model you’d like to use.
promptTemplate	Object	Optional	An override for the prompt template used in the service. It has one property template, which expects a string.
system	String	Optional	System prompt to guide the OCR process.

HTTP Request

POST /domo/ai/v1/image/text HTTP/1.1
Content-Type: application/json
Accept: application/json

Request Body

The required fields in the request body are the input string and the image object. You can provide additional properties in the request body to further customize the OCR service. If you choose to use the promptTemplate property to override the default prompt template, you’ll need to use ${input} in the template string to pass through the input of your prompt.

{
  "input": "<input prompt>",
  "image": {
    "mediaType": "image/png",
    "type": "base64",
    "data": "<base64-encoded-image - Some formats may be restricted, such as HEIC or HEIF>"
  },
  "model": "domo.domo_ai.domogpt-chat-medium-v1.1:anthropic",
  "promptTemplate": {
    "template": "${input}"
  },
  "system": "You are an AI assistant tasked with performing Optical Character Recognition (OCR) on an image. ..."
}

HTTP Response

Returns the prompt used and the response from the model. The response will include the transcribed text from the image, typically within <answer> tags as specified in the system prompt.

HTTP/1.1 200 OK
Content-Type: application/json;charset=UTF-8

{
  "prompt": "return the text in the image, ensuring new lines are preserved",
  "choices": [
    {
      "output": "<The output text from the image, e.g., 'Hello World'>"
    }
  ],
  "modelId": "domo.domo_ai.domogpt-chat-medium-v1.1:anthropic",
  "isCustomerModel": false
}