ixp
latest
false
  • Overview
    • Introduction
    • Extracting data from unstructured documents
    • Building and deploying models
  • Model building
    • Overview
    • Uploading sample documents​
    • Reviewing predictions
    • Validating extraction predictions
    • Configuring the model
  • Model validation
  • Model deployment
  • Frequently asked questions
    • Frequently asked questions
Important :
Communications Mining is now part of UiPath IXP. Check the Introduction in the Overview Guide for more details.
UiPath logo, featuring letters U and I in white
PREVIEW

Unstructured and complex documents user guide

Last updated Jul 2, 2025

Frequently asked questions

Access and availability

How do customers and partners get access to the IXP Unstructured and complex documents capability?

Customers and partners can register their interest on the Insider Portal, since the Unstructured and complex documents capability is currently available in the preview program. We regularly review registrations, and the selected customers and partners are notified to join the preview program. Make sure to provide as much detail about the use case as possible.

When will the capability be officially Generally Available?

The target date to officially release the Unstructured and complex documents capability is late July, 2025. Selected customers can participate in a controlled General Availability during the remainder of the preview period. All controlled Generally Available use cases will need to be approved by the Product team and supported by their respective account teams.

Will the capability be available for Community users?

The capability is now available to Community users, who only have access to a restricted functionality, and all tenants or projects are subject to pre-defined design-time and run-time limits.

What are the IXP deployment options? Is the capability available on-premise?

Currently, the IXP Communications Mining and Unstructured and complex documents capabilities are only available in the UiPath® Automation Cloud™.

Document Understanding is available in Automation Cloud, Automation Suite, and Standalone deployments. For more information on Document Understanding deployment options, check Choosing the deployment type.

Is the capability available in all regions?

Currently, the Unstructured and complex documents capability is only available in AMER and EMEA tenants.

What languages are supported in the capability?

The official release for the Unstructured and complex documents capability includes English only. However, users can upload documents and build models in any language that is supported by the selected model within IXP.

What file formats are supported in the capability?

Currently, the capability supports PNG, JPG or JPEG, PDF, and TIF or TIFF file formats.

Data and security

Does IXP use specialized or generative AI?

Communications Mining uses specialized AI and generative AI features, where all generative AI features can be disabled if required.

Document Understanding uses specialized AI and generative AI features, where all generative AI features can be disabled if required.

Generative Extraction for Unstructured and complex documents uses generative AI via third-party LLMs. It is not possible to use this capability without generative AI.

Are CommPath and DocPath available in IXP?

CommPath is available in Communications Mining when using the Generative Extraction feature.

DocPath is available in Document Understanding, only in the Modern experience. available in the Document Understanding Classic experience.

Neither is available in the Generative Extraction for Unstructured and complex documents capability.

Does IXP go through the AI Trust Layer (AITL)?

Yes, all IXP capabilities that use third-party LLMs go through the UiPath AITL. CommPath and DocPath are the proprietary models of UiPath and do not go through the AITL.

Which LLMs are used by the capability?

Currently, only the GPT4o vision model is available in the IXP Unstructured and complex documents capability, but additional LLMs will be added in future releases.

Does the capability enable BYO (bring your own) subscription?

Not currently, but BYO subscription will be supported in future releases.

Migration

Can I migrate my use case from Document Understanding Classic experience to the IXP Unstructured and complex documents capability?

Not yet, but the import and export functionality will be available in future releases.

Can I migrate my use case from Document Understanding Modern experience to the IXP Unstructured and complex documents capability?

Not yet, but the import and export functionality will be available in future releases.

Can I migrate my use case from Document Understanding Generative Extractor activities to the IXP Unstructured and complex documents capability?

Not yet, but the import and export functionality will be available in future releases.

Design-time

Are there any design-time limitations for the capability?

Design-time limitations include:

  • Number of fields per field group: up to 32 fields.
  • Number of characters per field name: up to 32 characters (to be increased in future releases).
  • Number of characters per prompt instruction: up to 2048 characters. This is applicable to all prompts, including overall instructions, field group instructions, field instructions, and field type instructions.

Can I create generic fields in the documents capability?

No, there is no concept of generic fields in IXP. All fields need to belong to a field group. Each field group represents a separate table, where fields mean table columns. However, it is possible to define a parent-level field group representing a document type, and add more general fields as a table under this field group.

Do IXP Unstructured and complex documents models improve performance with labelling or annotations?

No, the model does not yet learn directly from user annotations. The way to currently improve the performance of your model is to modify your prompt instructions or model configuration. We are researching different solutions, including in-context learning, instruction feedback improvements, and fine-tunable models.

Why do I need to annotate documents if it does not improve the performance of the model?

It is still important to validate and annotate documents to establish ground truth for the performance evaluation of the model, which is captured in the built-in performance metrics in the Measure page.

How many documents do I need to upload to test my prompt instructions?

You should upload a representative sample of your documents to cover all possible variations. This should be at least 20-30 different documents. You are recommended to review the initial predictions on 2 to 3 documents and decide whether you need to modify your prompt instructions before gradually increasing the test sample to 10, 20, 30, X documents, where this will be an iterative process.

How many documents do I need to annotate to get reliable performance metrics?

You are recommended to validate at least 10 documents or 10 examples, or both, for each field, so that the model can provide you with meaningful validation statistics. This is the minimum number of annotations and will depend on the specific use case requirements.

Run-time

Are there any run-time limitations for the Unstructured and complex documents capability?

Run-time limitations include:

  • Number of pages per document: documents up to 50 pages.
  • Token limits: long documents or extended tables with more than 300 cells will likely reach the token limit.

In a pre-General Availability release, these limits should both be significantly increased. The increased limit for a document will be a product of the number of pages and the number of field values that need to be extracted. If the combination of these two is high enough, the maximum limit will be reached, but the expectation is to support documents with 100-200 pages and high numbers of field values requiring extraction.

Note:

There is no hard limit on field values to be extracted, as it will depend on the document. However, if the token limit is reached, an indicator is provided in the API response for any run-time document predictions in which the occurrence and extraction confidence for an extraction will both be returned as zero.

Currently, if more than 50 pages are sent, the capability only uses the first 50 pages. If fewer than 50 pages are used, but the token limits are hit, it returns empty predictions.

Are Unstructured and complex documents models consumed via robots or standalone Document Understanding APIs?

The Unstructured and complex documents models can be consumed via both robots and standalone Document Understanding Automation Cloud API.

How can I consume Unstructured and complex documents outputs?

Model outputs can be consumed via APIs, Document Understanding Activities, or IntelligentOCR Activities. Make sure to:

  • Install the IOCR package v6.23.2-preview or newer and select the Document Understanding Project Extractor activity
  • Install the Document Understanding package v2.13.2-preview or newer and select the Extract Document Data activity

Does the IXP Unstructured and complex documents capability provide confidence scores? How are they calculated?

Yes, we provide confidence scores in the model response and in Action Center (run-time), but not in the IXP user interface (design-time). We use the token probabilities returned by the LLM to estimate field and extraction level confidences.

Can the Unstructured and complex documents model outputs be consumed from hybrid deployments, that are on-prem Orchestrator or Automation Suite and Cloud IXP model?

Yes, this is supported via External Apps. For more details, check OAuth apps.

Can IXP Unstructured and complex documents model outputs be consumed from different organizations or tenants, that are Production, Development, or UAT?

Yes, this is supported via External Apps. For more details, check OAuth apps.

Does the validation station work with the Unstructured and complex documents capability?

Yes, it works the same way as Document Understanding Modern or Classic projects. However, all extractions need to be validated as tables.

Does the Unstructured and complex documents model learn based on HITL validation from Action Center?

No, currently there is no feedback loop between Action Center and Unstructured and complex documents models. This is planned to be supported in future releases alongside fine-tunable models.

Maintenance and monitoring

Does the capability offer any monitoring tools? Are these different than Document Understanding Modern projects? Is there a dependency on Insights?

Currently, the Unstructured and complex documents capability does not provide any monitoring capabilities, such as the Monitor page in Modern Projects. Additional details on monitoring capabilities will be shared in future releases.

Can we monitor Platform Unit or AI Unit consumption for the Unstructured and complex documents projects?

The IXP AI unit consumption is being included in the next preview and subsequent General Availability release of the AI Units dashboards.

Can we auto-detect performance decay, for example, data drift, lower F1 scores, and so on?

For now, we do not provide any model performance notifications or alerts.

How often do users need to update their models or prompt instructions?

This depends on each use case, but you should refine your prompt instructions whenever your extractions are underperforming.

Was this page helpful?

Get The Help You Need
Learning RPA - Automation Courses
UiPath Community Forum
Uipath Logo White
Trust and Security
© 2005-2025 UiPath. All rights reserved.