AI Agent Studio FAQs
- Updated: 2025/09/26
Refer to this FAQ for general queries on the behavior and functionality of creating Model connections, AI Skills, running automations using the AI Skills package, and tracking and monitoring model interaction data within automations for security and governance.
General FAQs
- What are the supported foundational models?
- Currently, AI Tools supports Standard, Fine tuned and RAG
foundational models from hyperscaler vendors such as Amazon Bedrock, Google Vertex AI, Azure OpenAI, OpenAI and others.
These are Standard and Fine tuned models supported for each
vendor:
- Amazon Bedrock: Titan Text G1 - Express, Titan Text G1 - Lite, Anthropic Claude 3.5 Sonnet, Anthropic Claude 2.1.
- Automation Anywhere: Claude 3.5 Sonnet, GPT-4o. Note: The licensing and usage of these models are managed by Automation Anywhere through a credit-based system, so you do not need to provide your own API keys.
- Azure OpenAI: GPT 3.5 Turbo, GPT-3.5 Turbo (16k), GPT-4, GPT-4 (32k), GPT-4.1, GPT-4o, GPT-4o mini.
- Google Vertex AI: Anthropic Claude 3.5 Sonnet, Gemini 1.0 Pro, Gemini 1.5 Flash, text-bison, text-bison-32k (latest), text-unicorn.
- OpenAI: GPT 3.5 Turbo, GPT-3.5 Turbo (16k), GPT-4, GPT-4 Turbo Preview, GPT-4.1, GPT-4o, GPT-4o mini.
These RAG models are supported for each vendor:- Amazon Bedrock: Amazon Titan Text Premier, Anthropic Claude 3 Haiku v1, Anthropic Claude 3 Sonnet v1, Anthropic Claude Instant v1, Anthropic Claude v2.0, Anthropic Claude v2.1.
- Automation Anywhere: Supports all foundational models that Enterprise Knowledge supports.
- Azure OpenAI: GPT 3.5 Turbo, GPT-3.5 Turbo (16k), GPT-4, GPT-4 (32k), GPT-4o.
- Google Vertex AI: Gemini 1.0 Pro 1, Gemini 1.0 Pro 2 , Gemini 1.5 Flash 1.
Important: To connect to external generative AI models in AI Agent Studio, you will need to bring your own licenses (BYOL), such as API keys, access keys, or tokens, from the respective model providers. - What is a Grounded Model connection?
-
A Grounded Model connections uses a Retrieval Augmented Generation (RAG) capability to produce responses that are grounded in a specific knowledge base or data store. This ensures the information is accurate and contextually relevant.
This feature is available for the following vendors:
- Automation Anywhere: Create a Grounded by Enterprise Knowledge connection to ground responses in your knowledge base.
- Azure OpenAI: Create a Grounded by AI Search connection to use an AI Search index as a data source.
- Amazon Bedrock: Create a Grounded by knowledge base connection to use an Amazon Knowledge Base.
- Google Vertex AI: Create a Grounded by data store connection to use a Google Data Source.
- What is a Fine tuned Model connection?
- A Fine tuned Model connection can be created by customizing the supported foundational models via testing and fine-tuning and making them available for use by connecting to the created AI Skills.
- Does AI Agent Studio allow customers to connect to custom models from other hyperscalers or models hosted on private, customer infrastructure?
-
Yes, you can define and connect to your own custom models. For more details on this functionality, refer to the Custom model definitions.
- What is the license requirement?
- Access to AI Agent Studio is included with the Enterprise Platform that enables AI Agents. Credit consumption is governed by Automation AI credits. For more information, refer to Enterprise Platform and Licensing model for AI Agents
- What type of data is stored and for how long when I enable data logging?
- When data logging is enabled, all model interactions, including prompts, model responses, and parameters, are stored during automation or prompt executions. This data is stored within your environment and is used to offer detailed insights into AI usage in automations.
- What are the supported hyperscaler vendor models for the new AI Governance code analysis rule?
- This is a list of supported models for each hyperscaler vendor:
- Amazon Bedrock: Jurassic-2 Mid, Jurassic-2 Ulta, Claude Instant v1.2, Claude v1.3, Claude v2.1 (other supported versions), Titan Text G1 - Lite, and Titan Text G1 - Express (other supported versions).
- Google Vertex AI: chat-bison (latest), chat-bison-32k (latest), chat-bison-32k@002, chat-bison@001, chat-bison@002, codechat-bison, codechat-bison-32k, codechat-bison-32k@002, codechat-bison@001, codechat-bison@002 (other supported versions), text-bison (latest), text-bison-32k (latest), text-bison-32k@002, text-bison@001, text-bison@002, text-unicorn@001, code-bison (latest), code-bison-32k@002, code-bison@001, code-bison@002, code-gecko@001, code-gecko@002, code-gecko (other supported versions), and Gemini Pro.
- OpenAI: gpt-3.5-turbo (default), gpt-3.5-turbo-16k, gpt-4, gpt-4-32k (other supported versions), text-davinci-003, text-davinci-002, davinci, text-curie-001, curie, text-babbage-001, babbage, text-ada-001, and custom models.
- Is the data for prompt query and prompt-tuning, displayed in the AI Governance logs stored in Cloud?
- AI Governance logs captures the model interactions when automations are executed. The prompt and model responses exchanged with the hyperscaler models can be audited.
- What happens to the data that is sent to a third-party model that processes prompts and generates an output?
- The Audit log provides details of the prompts and model responses that were sent to the hyperscaler models. Automation Anywhere does not have access to any third-party models you have subscribed to and the data stored or processed by them. We recommend referring to their relevant documentation for details on how your data is processed and stored by them.
- Will a third-party, such as Azure OpenAI, be able to access my data and use it?
- Automation Anywhere does not have any control over what data is used or stored by third-party models that you subscribe to, when you send queries to their models. We recommend referring to their relevant documentation for details on data processing and storage.
- What is the data retention policy for stored data?
- We adhere to a consistent data retention policy for audit logs. You can review the policies here: Automation 360 FAQ.
- Can I use SIEM integrations to forward the stored logs?
- Yes, stored data can be forwarded to an integrated SIEM system configured in the Control Room.
- What security measures are implemented to protect stored prompts?
- All logged data is stored within the customer's production environment, never leaving the protected boundaries. This data is used solely for analytics to provide insights into AI usage within automations. Prompts and model responses are encrypted using industry-standard algorithms, and log data is securely transmitted via TLS. Access to view the prompts and responses is controlled through role-based access control.
- Can the text prompts and model responses be exported or printed?
- To safeguard privacy, prompts and model responses cannot be exported or printed from the Control Room. However, they can be forwarded to a SIEM through a SIEM integration.
- Do I need additional licenses for data logging?
- Data logging and AI Governance is part of the Enterprise Platform license, available through AI Agent Studio.
For details about the Enterprise Platform license, see Enterprise Platform.
Known product behavior
- In Automation 360 v.34 release, AI Governance Audit logs are not supported for API Tasks. This is specific to the API Tasks that are run in real-time mode (for attended automations).
- Users must have the View Settings permission to edit an AI Skill. Otherwise, an error message is displayed and users will not be able to edit the AI Skill.
- In the AI Skill editor, when you try to update the title of an existing AI Skill without an associated Model connection, Prompt-text, or defined Prompt Input, the system displays an error message stating a validation failure and fails to save the update. The AI Skill should take the update and save the change.
- Some of the models do not return Token consumption data hence the AI Governance logs display a value of '0' in the log detail fields.
- If you delete a Model connection that is connected to an AI Skill, you will get an error while executing the Task Bot that has been configured to use that AI Skill.
- If you delete a Model connection and create a new one with the same name, the AI Skill will reference the new version. However, if the underlying LLM models are different the outcome could be different.
- For Microsoft Azure connections:
- The Model connection configuration requires a user to
provide a Deployment ID which is the name of the deployment within
the Microsoft Azure portal. As this deployment is
mapped to a specific foundational model, users should ensure that
they select the correct Model connection model that
maps to the Microsoft Azure deployment model.Note: You would see a warning message if these values do not match.
- Users should be aware that altering the connection details of an existing Model connection would save incorrect input details and will not automatically trigger the Test connection functionality. This could result in AI Skills, Task Bots, or API Tasks using the Model connection, to break during execution. If you make any changes to a Model connection, ensure you test the updated connection so the Bots using this Model connection do not fail during automation execution.
- The Model connection configuration requires a user to
provide a Deployment ID which is the name of the deployment within
the Microsoft Azure portal. As this deployment is
mapped to a specific foundational model, users should ensure that
they select the correct Model connection model that
maps to the Microsoft Azure deployment model.
- For AI Governance, there is a dependency on Bot Agent 22.60.10 and later for generating audit logs. If your Bot Agent is lower than 22.60.10, you will get a notification asking you to install a compatible Bot Agent version to ensure the audit logs for all model interactions are logged successfully in the Control Room. As an Admin user, you will receive a notification if any device registered to the Control Room is below the recommended Bot Agent version.
- If you encounter a run-time error during an automation execution, based on the system-displayed error message, 'Unable to publish AI Governance audit logs. Please make sure that the Bot Agent is at least 22.60.10 or later. Error: <java.lang.NoSuchMethodError error message>' we recommend updating to the latest available Bot Agent version.
- If you are unable to see any Audit log data, we first recommend checking for relevant custom role permissions. If you have the required permissions, then we suggest making sure you are using the latest available Bot Agent version. We recommend using Bot Agent 22.60.10 and later.
- For Audit log and Bot execution features to work smoothly, we recommend using compatible Control Room and Bot Agent versions. See Assign roles and permissions to enable AI Governance to view the version compatibility matrix.
- The Automation Command Center (ACC) updates and refreshes data every five minutes. However, the aggregated data takes 15 to 20 minutes for the data roll-up after a complete Control Room refresh.