Bring your own subscription (BYOS)

Configure IXP to route extraction requests through your own Google Vertex AI or Microsoft Azure OpenAI subscriptions.

概要

注:

この機能はプレビューで利用可能です。

BYOS lets you run IXP on your own large language model (LLM) subscriptions instead of the UiPath®-hosted models.

Because traffic routes through your own subscription of the selected third-party LLM, your data stays within your compliance boundary. This supports data residency and compliance requirements that a UiPath-hosted subscription cannot meet on its own. Projects that run entirely on your own subscriptions are also eligible for discounted consumption rates. For more details, refer to BYOS discounting.

Requests still pass through the AI Trust Layer LLM gateway, so policy enforcement and audit logging continue to apply to every call. For how the gateway routes requests, and for the difference between BYOS and Bring your own model (BYOM), refer to Configuring LLMs.

対応機種

BYOS covers every model that IXP offers. You need to bring a subscription for at least one of them, not for all of them. Each model that you bring needs its own entry in your AI Trust Layer LLM configuration.

Model family	Model version in IXP	プロバイダー	コネクタ
Gemini	2.5 Flash	Google Vertex AI	Google Vertex
Gemini	2.5 Pro (preview)	Google Vertex AI	Google Vertex
Gemini	3.1 Pro Preview (preview)	Google Vertex AI	Google Vertex
Gemini	3.5 Flash Lite (preview)	Google Vertex AI	Google Vertex
GPT	gpt-4o	Microsoft Azure OpenAI	Microsoft Azure OpenAI
GPT	gpt-5.1-2025-11-13 (preview)	Microsoft Azure OpenAI	Microsoft Azure OpenAI
GPT	gpt-5.4-2026-03-05 (preview)	Microsoft Azure OpenAI	Microsoft Azure OpenAI

Preview models are in scope for BYOS on the same terms as the generally available ones. If you bring a subscription for a preview model, the capabilities that use it run on your own subscription and count towards discount eligibility.

Model coverage and alternatives

You can configure multiple BYOS models at the same time. IXP resolves each LLM-based capability against the models that you configured. The LLM-based capabilities are pre-processing, extraction, and model-based attribution.

If a capability needs a model that your subscriptions do not cover, IXP offers the following alternatives:

• A UiPath-hosted model, if your AI governance policy permits it.
• A non-LLM alternative, such as Table model - mini for pre-processing, or Rules-based for attribution.

The more models you bring, the more IXP capabilities run on your own subscription. Some pre-processing and post-processing features are Gemini-based and only have non-LLM alternatives if you do not bring a Gemini subscription.

BYOS discounting

Projects configured with BYOS are eligible for a 30% discount on typical IXP consumption rates.

The discount applies at the project level, and only if the project uses no UiPath-hosted LLM subscription at all. A project qualifies only when none of the following run on a UiPath-hosted model:

• LLM-based pre-processing, such as Table model.
• The extraction model.
• Model-based attribution.

If any one of these capabilities runs on a UiPath-hosted LLM subscription, the entire project is billed at standard consumption rates. There is no partial or pro-rated discount.

When you plan your configuration, consider the following:

• Selecting a non-LLM alternative, such as Table model - mini or Rules-based attribution, does not disqualify a project. These alternatives do not consume a UiPath LLM subscription.
• Falling back to a UiPath-hosted model for any of the three previously mentioned options disqualifies the project.
• To guarantee eligibility, disable the first-party model toggles in your AI governance policy, as described in Create an AI governance policy. Only your BYOS subscription is then used, so a project cannot fall back to a UiPath-hosted model.
• Eligibility is evaluated per project, so the same tenant can contain both discounted and standard-rate projects.

前提条件

Before you configure BYOS for IXP, make sure you have:

• Access to AI Trust Layer in your UiPath tenant.
• Access to UiPath Integration Service.
• A valid subscription to Google Vertex AI, Microsoft Azure OpenAI, or both.
• Credentials to create a connector in Integration Service.

1. Create a connection in Integration Service

Create a connection to each LLM provider that you intend to use before you configure it in AI Trust Layer.

1. Navigate to Integration Service in your UiPath tenant.
2. Go to My Workspace or Shared, depending on where you want the connection to be available.
3. Select the Connectors tab, then search for your provider:
   • For Gemini models, search for vertex and select Google Vertex.
   • For GPT models, search for azure open and select Microsoft Azure OpenAI.
4. Select the connector to open the connection form.
5. Complete the fields required by your provider.
6. [接続] を選択します。
7. If you are bringing models from both families, repeat steps 2 to 6 for the other provider.

Remember which folder you saved the connection to. You select it in the Connections Folder field when you add the LLM configuration.

2. Configure your models in AI Trust Layer

Configure your LLMs in AI Trust Layer as follows:

1. Navigate to AI Trust Layer, then LLM configurations.

2. [設定を追加] を選択します。

3. 次のフィールドに入力します。

   • Tenant - The tenant you want to configure.
   • Product - IXP
   • Feature - Unstructured & Complex Documents
   • Connections Folder - The folder where you saved your Integration Service connection, for example Shared.

4. Under Model Configuration, add one entry for each model that you are bringing. Add at least one entry, and as many as you have subscriptions for. Models that share a provider, such as several Gemini models, can reuse the same Integration Service connection.

   • For Gemini models, fill in the following fields:

     • LLM Name - The identifier of the model.
     • API Type - Gemini Generate Content
     • Connector - Google Vertex
     • Connection - The Google Vertex connection you created.
     • LLM identifier - Enter the identifier of the model manually.

   • For GPT models, fill in the following fields:

     • LLM Name - The identifier of the model.
     • API Type - OpenAI Chat Completions
     • Connector - Microsoft Azure OpenAI
     • Connection - The Microsoft Azure OpenAI connection you created.
     • LLM identifier - Enter the identifier of the model manually.
   注:

   There is a known issue where the autocomplete for the LLM identifier dropdown might not work correctly in this preview. Make sure you enter the identifier directly instead of relying on the suggestions.

5. Select Test configuration to validate that AI Trust Layer can reach each configured LLM provider with the credentials and identifiers you provided.

6. On the Configuration validated confirmation, select Done.

7. [保存] を選択します。

Model identifiers

Enter the following identifiers for the models that you configure:

Model version in IXP	識別子
2.5 Flash	gemini-2.5-flash
2.5 Pro (preview)	gemini-2.5-pro
3.1 Pro Preview (preview)	gemini-3.1-pro-preview
3.5 Flash Lite (preview)	gemini-3.5-flash-lite
gpt-4o	gpt-4o-2024-08-06
gpt-5.1-2025-11-13 (preview)	gpt-5.1-2025-11-13
gpt-5.4-2026-03-05 (preview)	gpt-5.4-2026-03-05

Each model that you add must also be deployed and available in your provider subscription. For example, an entry for gpt-5.4-2026-03-05 is validated only if your Azure OpenAI resource has that deployment.

Troubleshooting test configuration

If testing the configuration fails, check the following:

• Your credentials in Integration Service are correct, and the connection is active.
• The correct folder is selected in the Connections Folder field.
• The LLM identifier matches the value for your model exactly.
• Each model entry that you added has its own valid connection. Validation fails if any single model in the configuration fails.
• Each model that you added is deployed in your provider subscription.

3. Create an AI governance policy

After you save your LLM configuration, you need to set up an AI governance policy to control which models are permitted in your tenant. Create the policy as follows:

1. On the AI Trust Layer landing page, select the AI Governance tab.
2. Select Add Policy.
3. Enter a policy name, for example IXP_BYOS_Policy, and optionally a description.
4. Select the Models tab.
5. Set the Anthropic, Open AI, and Gemini toggles. This setting determines whether your projects can fall back to UiPath-hosted models, and therefore whether they qualify for BYOS discounting:
   • Enabled: UiPath-hosted versions of those models remain available alongside your BYOS configuration. This gives you broader capability coverage, but any project that ends up using a UiPath-hosted model for pre-processing, extraction, or model-based attribution loses the discount.
   • Disabled: only your BYOS subscription is used. Capabilities that your models do not cover fall back to their non-LLM alternatives, and your projects remain eligible for the discount.
6. Select Submit to apply the model settings.
7. Select Save to save the policy.

For more details, refer to Settings for AI Trust Layer policies.

4. Assign the policy to your tenant

After you save the policy, deploy it to the tenant where IXP runs:

1. On the AI Governance tab, select Manage Deployments.
2. Under the Tenants tab, locate the tenant you want to apply the policy to, then select the edit icon next to it.
3. Expand the No License section.
4. In the AI Trust Layer row, select the policy you created from the Product Policy dropdown.
5. [保存] を選択します。

5. Verify your subscription

Once your configuration and policy are saved, verify that your subscription is active:

1. In IXP, create a project, or open an existing one.
2. [モデルの設定] を選択します。
3. Under Model family and Model version, select your model and version. What you can select depends on the models that you configured. For details, refer to the following sections.
4. Configure the preferred settings under Intelligent pre-processing, Attribution, and Advanced options.
5. [保存] を選択します。

Model families that you did not configure are shown as unavailable, unless first-party models are enabled in your governance policy. Any capability that depends on an unavailable family offers its alternative instead.

注:

The Model version dropdown lists every version that IXP supports, not only the versions that you brought a subscription for. If you select a version that your subscription does not cover, that capability falls back to the UiPath-hosted model, if your governance policy permits it, and the project no longer qualifies for BYOS discounting.

For the full set of model settings, refer to Model configuration.

If you configured Gemini models only

• The GPT model family is unavailable.
• Under the Gemini family, you can select the versions 2.5 Flash, 2.5 Pro, 3.1 Pro Preview, and 3.5 Flash Lite. The versions that you brought a subscription for run on BYOS.
• Gemini-based pre-processing and attribution options remain available.

If you configured GPT models only

• The Gemini model family is unavailable.
• Under the GPT family, you can select the versions gpt-4o, gpt-5.1-2025-11-13, and gpt-5.4-2026-03-05.
• Warnings indicate that certain pre-processing and post-processing options are unavailable. These options depend on Gemini and cannot be used without a Gemini subscription. They are grayed out, and you can use the following alternatives instead:

Affected setting	Where it appears	Alternative
表モデル	インテリジェントな前処理	表モデル - mini
モデルベースの属性付け	属性付け	ルール ベース

Selecting these alternatives keeps the project fully on your own subscription, so the project remains eligible for BYOS discounting.

If you configured models from both families

• You can select both the Gemini and the GPT model families.
• Gemini-based pre-processing and attribution options are available and run on your own subscription.
• If you brought every model that the project needs, no capability falls back to a UiPath-hosted model, and the project qualifies for BYOS discounting on any combination of settings.

Confirming discount eligibility

Before you save the model configuration, check the following settings:

機能	Qualifies for the discount when set to
LLM-based pre-processing (Table model)	A model version that you brought a subscription for, or Table model - mini
抽出モデル	A model version that you brought a subscription for
属性付け	Model-based on a model version that you brought a subscription for, or Rules-based

注:

If any of these capabilities is set to a UiPath-hosted model, the project is billed at standard rates.

Eligibility is checked per model version, not per model family. For example, bringing Gemini 2.5 Flash does not cover a project that is configured to extract with Gemini 3.1 Pro Preview.

6. Test the configuration

Once you save the model configuration, you can validate the full extraction pipeline using BYOS as follows:

1. Create your taxonomy to define the fields you want to extract.
2. Upload documents to your project.
3. Run extraction.

If you receive predictions, IXP is routing requests through your configured LLM subscription.