Free cookie consent management tool by TermsFeed

AI Services

On This Page

Purpose

The AI Services page in the Admin Console gives you more control over which generative AI models are used in your Appian applications. As AI providers release new and more powerful models, this screen allows you to adopt them in a secure and controlled way.

Access

To access the AI Services page, go to the Admin Console. Then, in the left navigation menu, under Integration, click AI Services.

Settings

The AI Services screen gives you more control over which generative AI models are used in your Appian applications. As AI providers release new and more powerful models, this screen allows you to adopt them in a secure and controlled way.

These settings directly impact Appian AI capabilities that leverage large language models. To use newer models, you must first enable them here. This is a critical step for migrating any existing capabilities that use legacy models, like Sonnet 3.5, which will be deprecated.

The AI Services screen contains the following tabs:

  • Appian
  • Google Cloud and Microsoft Azure: Use these tabs to configure connections to your organization's accounts with third-party AI service providers. The Appian features that relied on both of these tabs are deprecated.

Select AI models and inference profiles

System administrators can select models and inference profiles on the Appian tab.

  • For sites in a US region, a US inference profile is enabled by default.
  • For FedRAMP sites, AI models are automatically configured based on service agreements with Appian. You cannot select a model in these environments.

To configure a model and profile:

  1. In the Admin Console, navigate to the AI Services page.
  2. On the Appian tab, on the card for the model you want, use the Selected Profile dropdown to choose a profile.

    Note:  Selecting an inference profile outside your site's primary region may introduce data latency.

  3. Click SAVE CHANGES.

Tip:  To deactivate a profile, select the None option from the top of the list.

For more information about choosing generative AI models for your AI capabilities, see the AI Services page.

Google Cloud

Note:  Customers who wish to use Google services in Appian will need to bring their own Google Cloud credentials to do so.

In order to use certain Document Extraction features, you need to enter your Google Service Account key information in the Admin Console > AI Services page.

The required information is in the Google Service Account Key file, which is generated when the service account key is created. These keys will be used for all instances of the Doc Extraction smart services.

How to get a Google service account key file

Contact your Appian administrator for the Google Service Account Key JSON file from the License Management section of Appian Community.

Otherwise, to create the Service Account Key file:

  1. Ensure APIs are enabled - in the Google Cloud Console, navigate to APIs & Services > Library - see Google Cloud docs on enabling APIs.
    1. Google Cloud Storage JSON API
    2. Cloud Document AI API
  2. Generate a Google Cloud Storage Bucket - see the Google Cloud documentation on creating storage buckets
  3. Create Google Service Account with the following permissions:
    • storage.objects.create
    • storage.objects.list
    • documentai.processors.create
    • documentai.processors.list

    In the Google Cloud Console, navigate to APIs & Services > Credentials page - see the Google IAM documentation for more information on creating and managing service accounts.

  4. Generate a Service Account Key - to use Google Service Account authentication, you will need a service account key for your service account. This file should be downloaded in JSON format. To learn about creating and managing service account keys, see the Google IAM documentation on creating and managing service account keys

The following Google Cloud Document AI keys are in the Google Service Account key JSON file that is generated when the service account key is created.

  • Project ID: Found in the Google Service Account file
  • Private Key ID: Found in the Google Service Account file
  • Private Key: Found in the Google Service Account file
  • Client Email: Found in the Google Service Account file
  • Client ID: Found in the Google Service Account file
  • Destination Bucket: The name of the Google Cloud Storage Bucket to upload the file to. The Google Service Account must have write access to this bucket. For example: storage-bucket-0000000000
  • Source Bucket: The name of the Google Cloud Storage Bucket to upload the file to. The Google Service Account must have read access to this bucket. For example: storage-bucket-0000000001
  • Compute Region: The region that Google will store and process the resources for your project. The following regions are currently supported: us, eu.
  • Test Connection: Uses the provided configuration information to connect to the target system. Returns success or error with details.

The Microsoft Azure tab allows you to provide credentials for the Azure OpenAI Service, which previously enabled certain Appian AI Copilot features such as generating an interface from a PDF in Appian 24.3 and earlier. After credentials were provided, AI Copilot capabilities displayed in your environment.

Note:  As of the Appian 24.4 release, generating an interface from a PDF is now powered by Appian's private AI. This enhancement eliminated the need for an external Azure OpenAI license.

To connect Appian to Azure OpenAI:

Tip:  These steps only apply if you are using Appian 24.3 and earlier.

  1. In the Azure Portal, create a resource. You'll use the resource name as your-resource-name in the endpoint you construct in step #3.
  2. Deploy a model.
    • For the Select a model field, choose gpt-4-32k (0314) or gpt-4 (0314). You'll use the name of the model as the deployment-id in the endpoint in step #3.
  3. Construct a Chat completions endpoint using your resource information. See Microsoft documentation for the format.
  4. Retrieve a key for your resource on the Keys and Endpoint page. You can use KEY 1 or KEY 2.
  5. In Appian, go to the Admin Console.
  6. In the Integrations menu, select AI Services.
  7. Go to the Microsoft Azure tab.
  8. Select the Connect to Azure OpenAI checkbox.
  9. Paste the endpoint URL into the Chat Completion Endpoint field.
  10. Paste the resource key into the Key field.
  11. Click TEST CONNECTION to see if the connection is successful. The success message includes information on any applicable Azure OpenAI service limits. Specifically, there is a limit to how many units of text Azure OpenAI processes per conversation, called the token limit. If the connection is unsuccessful, a message displays the reason why.
  12. Click SAVE CHANGES.

What data is sent to Azure OpenAI?

When you or an end user interacts with AI Copilot, certain information is sent to Azure OpenAI to process the request:

Action Information sent to Azure OpenAI
You upload a PDF to generate an interface. The content of the PDF.

For more information about Azure OpenAI data privacy, see the Microsoft documentation.

What is a token limit?

In the context of machine learning, a token is a unit of analysis. A model breaks down data into meaningful parts – tokens – before it begins processing that data. Language processing models break down paragraphs of text into their smallest units, but these units aren't always equivalent to words. Instead, a token could be part of a word. Therefore, the number of tokens won't always equal the number of words in the text analyzed.

For example, the model is given the sentence, "the quickest brown fox is jumping over the laziest dog." There are 10 words in this sentence. However, there could be 13 tokens if the model breaks longer words such as quickest into quick and est for analysis. Quickest is one word, but quick and est are two tokens.

Feedback