Configuring LLMs

Note:

LLM configurations is available on the following licensing plans:

• Flex: Advanced Platform, Flex Standard Platform.

The LLM configurations tab allows you to integrate your existing AI subscriptions while maintaining the governance framework provided by UiPath. You can:

• Add your own LLM: Use any LLM that meets the product's compatibility criteria. To ensure smooth integration, your chosen LLM must pass a series of tests initiated through a probe call before it can be used within the UiPath ecosystem.

Configuring LLMs preserves most of the governance benefits of the AI Trust Layer, including policy enforcement via Automation Ops and detailed audit logs. However, model governance policies are specifically designed for UiPath-managed LLMs. This means that if you disable a particular model through an AI Trust Layer policy, the restriction only applies to the UiPath-managed version of that model. Your own configured models of the same type remain unaffected.

When leveraging the option to use your own LLM or subscription, keep the following points in mind:

• Compatibility requirements: Your chosen LLM or subscription must align with the model family and version currently supported by the UiPath product.
• Setup: Make sure you properly configure and maintain all required LLMs in the custom setup. If any component is missing, outdated, or incorrectly configured, your custom setup may cease to function.
• Cost-saving: If your custom LLM setup is complete, correct, and meets all necessary requirements, you may be eligible for a Reduced Consumption Rate.

Setting up an LLM connection

LLM connections rely on Integration Service to establish the connection to your own models. You can create connections to the following providers:

• Amazon Web Services
• Azure Open AI
• Google Vertex
• Open AI
• Open AI V1 Compliant LLM – Use this option to connect to any LLM provider whose API follows the OpenAI V1 standard. For details, refer to the OpenAI V1 Compliant LLM connector documentation.

Note:

To configure Anthropic Claude models, use the Amazon Web Services connector in Integration Service. A direct Anthropic connector is not supported in Private Test Cloud.

To set up a new connection, follow these steps:

1. Create the Integration Service connection

1. Create a connection in Integration Service to your provider of choice. For connector-specific authentication details, see the Integration Service user guide.
   Note:

   To prevent unauthorized access, create the Integration Service connection in a private, non-shared folder.

2. Add a new LLM configuration

1. Navigate to Admin > AI Trust Layer > LLM Configurations.

2. Select the tenant and folder where you want to configure the connection.

3. Select Add configuration.

4. Select the Product and Feature.

5. Choose how you want to configure:

   • Add your own LLM – Add an additional LLM configuration managed entirely by you.

   Depending on the selected product, only one option may be available.

3. Configure the model

Set up the connection for Add your own LLM:

1. Folder – Select the folder where the configuration will be stored.
2. Displayed (LLM) name – Provide an alias for your LLM.
3. Connector – Select your connector (e.g., Microsoft Azure OpenAI).
4. Connection – Choose your Integration Service connection.
5. LLM identifier – Enter the identifier for your model.
   • For Azure-hosted models, enter the model identifier.
   • For AWS cross-region inference, enter the inference profile ID instead of the model ID.

4. Validate and save

1. Select Test configuration to check that the model is reachable and meets the required criteria.

   UiPath can confirm reachability, verifying the exact model used is your responsibility.

2. If the test is successful, select Save to activate the connection.

Managing existing LLM connections

You can perform the following actions on your existing connections:

• Check status – Verify the status of your Integration Service connection. This action ensures that the connection is active and functioning correctly.
• Edit – Modify any parameters of your existing connection.
• Disable – Temporarily suspend the connection. When disabled, the connection remains visible in your list but doesn't route any calls. You can re-enable the connection when needed.
• Delete – Permanently remove the connection from your system. This action disables the connection and removes it from your list.

Configuring LLMs for your product

Each product supports specific large language models (LLMs) and versions. Use the table below to identify the supported models and versions for your product.

You can connect your own LLM using one of the following providers: Amazon Web Services, Google Vertex, Microsoft Azure OpenAI, or OpenAI V1 Compliant. Follow the steps outlined in the previous section to create a connection.

Note:

File support: Some product features rely on the configured LLM endpoint to process uploaded files. When using custom LLM configurations, support for file formats depends on the provider, model family, model version, and API Type. Verify that the selected model supports the required file formats before enabling file-based features. For product-specific requirements, refer to the relevant product documentation — for example, Analyze Files for Agents.

The number of models you must configure depends on the product and feature:

• For features with a selectable model — where you choose which model to use — you can configure one or more models; unconfigured models continue to use UiPath-managed subscriptions.
• For features with a fixed model set — where the feature uses a predetermined set of models — all models must be configured for the feature to work; partial configuration is not valid.

Product	Feature	LLM provider	Version
Agents 1	Design, Evaluate & Deploy	Anthropic	anthropic.claude-3.5-sonnet-20240620-v1:0 anthropic.claude-3.5-sonnet-20241022-v2:0 anthropic.claude-3.7-sonnet-20250219-v1:0 anthropic.claude-3-haiku-20240307-v1:0
		Google	gemini-2.5-pro
			gemini-2.5-flash
		OpenAI	gpt-4o-2024-05-13 gpt-4o-2024-08-06 gpt-4o-2024-11-20 gpt-4o-mini-2025-04-14 gpt-4o-mini-2024-07-18
Autopilot	Generation	Google	gemini-2.5-flash-lite gemini-2.5-flash gemini-2.5-pro gemini-embedding-001
	Chat	Anthropic	anthropic.claude-haiku-4-5-20251001-v1:0 anthropic.claude-sonnet-4-6 anthropic.claude-opus-4-6-v1
		Google	gemini-2.5-pro gemini-2.5-flash gemini-3-flash-preview gemini-3-pro-preview gemini-3.1-pro-preview
Autopilot for everyone	Chat	Anthropic	anthropic.claude-3.5-sonnet-20240620-v1:0 anthropic.claude-3.7-sonnet-20250219-v1:0
		OpenAI	gpt-4o-mini-2024-07-18
Coded agents	Call LLM	Anthropic	anthropic.claude-3.5-sonnet-20240620-v1:0 anthropic.claude-3.5-sonnet-20241022-v2:0 anthropic.claude-3.7-sonnet-20250219-v1:0 anthropic.claude-3-haiku-20240307-v1:0
		Gemini	gemini-1.5-pro-001 gemini-2.0-flash-001
		OpenAI	gpt-4o-2024-05-13 gpt-4o-2024-08-06 gpt-4o-2024-11-20 gpt-4o-mini-2024-07-18 o3-mini-2025-01-31
Context Grounding	Embeddings	Gemini	gemini-embedding-001
		OpenAI	text-embedding-3-large
	Advanced ingestion	Gemini	gemini-2.5-flash
	DeepRAG	Gemini	gemini-2.5-flash
	Batch Transform	Gemini	gemini-2.5-flash gemini-2.5-flash-lite
	Batch Transform with Web Search	Gemini	gemini-2.5-flash gemini-2.5-flash-lite
GenAI Activities	Build, Test & Deploy	Anthropic	anthropic.claude-3.5-sonnet-20241022-v2:0 anthropic.claude-3.7-sonnet-20250219-v1:0
		Gemini	gemini-2.5-pro gemini-2.5-flash
		OpenAI	gpt-5-2025-08-07 gpt-5-mini-2025-08-07 gpt-5-nano-2025-0807
Healing Agent	Workflow Recovery
		Google	gemini-2.5-pro gemini-2.5-flash
		OpenAI	gpt-4o-2024-08-06
UI Automation	ScreenPlay	Anthropic	anthropic.claude-sonnet-4-5-20250929-v1:0
		Google	gemini-2.5-flash
		OpenAI	gpt-4.1-mini-2025-04-14 gpt-4.1-2025-04-14 gpt-5-2025-08-07 gpt-5-mini-2025-08-07 computer-use-preview-2025-03-11
	Semantic selectors	Google	gemini-2.5-flash
Test Manager	Autopilot Autopilot Search Find Obsolete Tests Generate Test Cases Import Test Cases Generate Reports Requirement Evaluation	Anthropic	anthropic.claude-3.7-sonnet-20250219-v1:0 (to be replaced with anthropic.claude-4.5-sonnet in March 2026)
		Google	gemini-2.5-pro gemini-2.5-flash
		OpenAI	gpt-4o-2024-11-20

¹ Agents requirements:

• Ensure your LLM supports:
  • Tool (function) calling – The model must be able to call tools or functions during execution.
  • Disabling parallel tool calls – If supported by your provider, the model should offer the option to disable parallel tool calls.
• When using custom models, Agents default to a 4096 token limit regardless of the model's true capacity, since UiPath cannot infer token limits for customer-defined deployments.