Edit

Agentic mode for document analysis (preview)

Agentic mode in Azure Content Understanding in Foundry Tools analyzes complex documents when an answer must be built from evidence instead of extracted from a single location. It can reason across a document, perform calculations, validate results, interpret visual information, and return structured fields that match your schema.

Important

API version 2026-06-01-preview is in public preview. Previews are provided without a service-level agreement and aren't recommended for production workloads. For more information, see Supplemental Terms of Use for Microsoft Azure Previews and the Microsoft Products and Services Data Protection Addendum ("DPA").

Important

Agentic mode is available with API version 2026-06-01-preview. The initial preview supports one input file per analysis request. The input file can contain a single document or multiple logically related documents (such as a contract, appendix, company rules, and so on).

When to use agentic mode

Agentic mode is the most advanced Content Understanding option for your most challenging document analysis problems. If standard extraction doesn't produce the result you need, try agentic mode when additional reasoning over the data in the input file could construct a better answer. Agentic mode applies the service's highest reasoning effort to help you get the most from its AI capabilities.

Use agentic mode for document scenarios that require one or more of these capabilities:

  • Multistep reasoning: Connect related information in different parts of a document to construct an answer.
  • Calculations: Derive values such as totals, differences, or reconciliations that aren't stated directly.
  • Validation: Check values for internal consistency or against conditions in your field descriptions.
  • Visual analysis: Use information in complex tables, figures, forms, and other visual elements as part of the analysis.
  • Structured output: Return the result as fields that conform to your analyzer schema.

For example, use agentic mode to reconcile values in a financial report, validate whether a document meets a set of requirements, or analyze how a contract and its amendments in the same file relate to each other.

Agentic mode can extract fields from a single document or reason across multiple logical documents contained in the input file.

For straightforward field extraction, use a standard document analyzer. Agentic mode isn't a replacement for human review in high-impact scenarios.

Select agentic workflow

Set config.workflow when you create a document analyzer by using API version 2026-06-01-preview:

  • Use "default", or omit workflow, to let the service select the standard or advanced workflow based on your analyzer configuration.
  • Use "agentic" to enable agentic mode.
{
  "description": "Calculate and validate totals in an invoice",
  "baseAnalyzerId": "prebuilt-document",
  "models": {
    "completion": "gpt-5.2"
  },
  "config": {
    "workflow": "agentic" // Agentic workflow selector
  },
  "fieldSchema": {
    "fields": {
      ...
    }
  }
}

The value you send is a creation-time selector. After the analyzer is created, config.workflow contains a resolved, versioned value. For example, "agentic" resolves to "agentic.2026-06-01-preview". For the resolution rules, see workflow.

Cost and latency

The resolved agentic.* workflow uses the advanced contextualization rate. Agentic mode also consumes more model tokens than a standard workflow and typically takes longer for the same document. Test with representative documents to determine whether the quality improvement meets your latency and cost requirements.

Note

Agentic mode typically requires approximately 400,000 tokens per minute (TPM) per analyzer job on the Foundry model deployment. Configure at least 400,000 TPM capacity to help avoid 429 rate-limit errors.

For more information, see Content Understanding pricing.

Preview limitations

The initial preview has these limitations:

  • Each analysis request supports one input file.
  • Agentic mode supports document analyzers only.
  • Fields that use the extract method aren't supported.
  • Using labeled samples to improve the analyzer isn't supported.

For other input limits, see Service quotas and limits.

Migrate from pro mode

Pro mode is only available in the 2025-05-01-preview API, which is now retired. Migrate to agentic mode in the 2026-06-01-preview API.

Keep the following differences in mind when you migrate:

  • Configure agentic mode through the 2026-06-01-preview API by setting config.workflow to "agentic". After creation, the service returns the resolved value "agentic.2026-06-01-preview". Agentic mode isn't a drop-in replacement for the 2025-05-01-preview pro mode configuration. Review your analyzer and field schema against the agentic mode documentation before you migrate.
  • The initial agentic mode preview supports one input file per analysis request. If your pro mode workflow relies on multiple input documents or reference data, confirm agentic mode supports your scenario before you migrate.

Related content

Use these resources to learn more about analyzers, document processing, and pricing:

  • Last updated on