> For the complete documentation index, see [llms.txt](https://docs.docbits.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.docbits.com/administration-and-setup/settings/document-processing/classification-and-extraction.md).

# Classification And Extraction

[AI Table](https://docs.docbits.com/end-user-and-partner-section/ai-table.md) · [Structured Output for Tables](https://docs.docbits.com/end-user-and-partner-section/ai-table/structured-output.md)

## Which setting works at which process step?

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-3f3eb3d022005400771b64c98adbdd521236d05e%2Fprocess-extraction-en-20261028.png?alt=media" alt="From file to recognized data: View document type, data source and extraction path separately."><figcaption><p>View document type, data source and extraction path separately. The diagram is schematic; the active settings of your organization determine the actual behavior.</p></figcaption></figure>

Classification answers the question **“What kind of document is this?”**. Extraction answers **“What values does it contain?”**. The settings on this page affect different points of processing and should not be understood as a single AI switch.

1. **Prepare input:** For PDFs, existing text or OCR can be used; supported electronic documents can provide structured values. An activated split can separate an incoming document based on the configured barcode or QR criteria.
2. **Determine document type:** Classification rules and models assign documents to a type and, if necessary, a subtype. This assignment is to be distinguished from the later field and table extraction.
3. **Extract values:** Tables can follow different paths depending on the configuration and applicable saved rules. Header fields have their own source indicators. The selected AI model alone therefore does not explain every row or field value.
4. **Trace the result:** In the validation screen, check which table tab contains the rows and which source indicator a header field has. If a processing step is missing, read the document logs before changing settings.

### Distinguish between rule path and AI path

Appropriate saved table rules can determine the table extraction for the relevant template or supplier. Without suitable rules, the AI path can be used when AI table extraction is activated. Individual columns can have their own AI options. An AI source for a header field does not mean that the entire table comes from the AI.

The [Which extraction path was used?](https://docs.docbits.com/overview-and-basics/faq/document-processing/which-extraction-path-was-used.md) diagnostic guide explains tabs, source identifiers and protocols. The page [AI model](https://docs.docbits.com/administration-and-setup/settings/document-processing/classification-and-extraction/ai-model.md) explains the organization standard and template-related overrides.

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-49355de4f1c5ba6e36433639bc216d5f97daa1d6%2Fdbdc-193-general-overview-en-20261030.png?alt=media" alt="Classification and Extraction with General, Classification Rules and AI Models, plus Document Splitting and Document Processing"><figcaption><p>General contains processing settings. AI Models is visible here because the module is enabled; its custom classification models are separate from the extraction-tier selector. The stored barcode pattern is obscured.</p></figcaption></figure>

## Overview

In the **Classification and Extraction** settings, you can:

* Control how uploaded documents are **split** and how many pages are processed
* Configure **amount formatting** for amount comparison
* Set up **table extraction**, including the AI model used per supplier
* Toggle processing of unsupported **ZUGFeRD** files
* Define **classification rules** that assign a document type automatically
* Review the **AI Models** and their trained labels

The page has **General** and **Classification Rules**, plus **AI Models** when the model and label module is enabled for your organization. Custom classification models and labels in **AI Models** are separate from the extraction-tier selector under **General → Extraction**.

| Tab                  | Contents                                                                                          |
| -------------------- | ------------------------------------------------------------------------------------------------- |
| General              | Document Splitting, Document Processing, Amount Formatting, Table Extraction, Electronic Document |
| Classification Rules | Regex rules that classify documents by source name                                                |
| AI Models            | Custom-trained models with their labels, document types and priority                              |

## Accessing Classification and Extraction Settings

Go to **Settings → Document Types → Classification and Extraction**.

## General

The **General** tab groups settings into collapsible sections. Most controls save when their value changes. A changed nonempty **Barcode Pattern** uses its validation/**Save** action, and a table-extraction version change requires confirmation. Review the organization and intended behavior before editing a control.

### Document Splitting

**Split** decides whether an uploaded document is split into several documents when a barcode appears on one of its pages. The dropdown offers:

* **Do not split uploaded documents** – the default; the document is processed as one file.
* **Split by Barcode/QR-Code** – the document is split at every detected barcode.

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-44b8361a416b1963bb967876620740dc8ec23a03%2Fdbdc-193-document-splitting-en-20261030.png?alt=media" alt="Document Splitting with barcode and QR-code splitting, selected barcode types and the obscured Barcode Pattern field"><figcaption><p>Review the split method and barcode types before changing them. The organization-specific pattern is obscured; no value has been changed.</p></figcaption></figure>

When **Split by Barcode/QR-Code** is selected, two more fields appear:

* **Barcode Type** – one or more barcode types to detect, for example *CODE-128* or *QR-CODE*.
* **Barcode Pattern** – a regex the barcode value must match before the split is triggered. Leave it empty to split on any barcode of the selected types. The pattern is validated before it is stored; an invalid expression is reported immediately.

After a successful split, the parent document is deleted and the pages continue as separate documents that share the same batch.

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-ed5487765ae7d399bc0ff89dfbca9d12c8680edc%2Fanatomy-barcode-split.svg?alt=media" alt="Isometric illustration of eight sheets in a tray with a scanner reading a barcode on sheet 6. Dividers stand in front of sheets 3 and 6, splitting the sheets into three documents, and one rubber band runs round the whole tray."><figcaption><p>Example with the pattern <code>^INV\d{6}$</code>: the codes on sheets 3 and 6 match and each starts a new document, the code on sheet 4 does not match and splits nothing, and the rubber band stands for the batch the resulting documents share.</p></figcaption></figure>

### Document Processing

**Pages to process** limits how many pages are OCRed, classified and extracted per document.

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-46aaf99c55b9bad267333a646b5296101e27542a%2Fdbdc-193-document-processing-en-20261030.png?alt=media" alt="Document Processing showing Pages to process set to 10 and the explanation that later pages remain stored and read-only"><figcaption><p>This example has a limit of 10 processed pages. The field explains that empty or 0 means all pages; the displayed value has not been changed.</p></figcaption></figure>

* Empty or **0** means all pages are processed.
* A number *N* limits OCR, classification and extraction to the first *N* pages. The complete document is still stored, displayed and exported; pages beyond the limit are shown read-only.

### Amount Formatting

Two switches control how amounts are compared against master data and purchase orders.

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-df393f5a7742b0468b048af5fe45504fecae6a3e%2Fdbdc-193-amount-formatting-en-20261030.png?alt=media" alt="Amount Formatting with Allow Rounding During Amount Comparison and Require Exact Match for Amount Comparison"><figcaption><p>The two amount-comparison switches are shown off in this example. Inspect the intended policy before changing either switch.</p></figcaption></figure>

* **Allow Rounding During Amount Comparison** – when enabled, a tolerance of ±0.5 is allowed. When disabled, ±0.05 applies.
* **Require Exact Match for Amount Comparison** – when enabled, amounts must match exactly, with zero tolerance. When disabled, ±0.05 applies.

Only one of the two switches can be active at a time; enabling one disables the other.

### Table Extraction

{% hint style="info" %}
**Prerequisites for a working table extraction**

* The document type has **table columns** (Settings → Document Processing → Document Types → [Table Columns](https://docs.docbits.com/administration-and-setup/settings/global-settings/document-types/table-columns.md)). Without columns there is nothing to extract into.
* **Table Extraction** or **AI Table Extraction** is switched on below, for the whole organization.
* The document has readable text: OCR ran, or E-Text is used for born-digital PDFs ([OCR Settings](https://docs.docbits.com/administration-and-setup/settings/document-processing/ocr-settings.md)).
* Training and AI models are **per supplier**. A trained table applies only to documents of the supplier it was trained on.
  {% endhint %}

You can extract tables from documents by enabling either **Table Extraction** or **AI Table Extraction**. A trained table (whether AI-based or manual) is always linked to a specific supplier.

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-fd17d5becea60c19df28e60fa4e8cd828f4566d7%2Fdbdc-193-table-extraction-en-20261030.png?alt=media" alt="Extraction controls for rule and AI table extraction, vision mode, structured extraction, hidden fields and the AI model selector"><figcaption><p>The extraction-tier selector is inside General → Extraction. The supplier-override table header is shown below; existing row data is obscured and no setting has been changed.</p></figcaption></figure>

**Table Extraction:** Activates rule-based table extraction. Tables are trained per supplier on the validation screen (*Go to table extraction view*).\
Learn more about training [here](https://docs.docbits.com/administration-and-setup/setup/document-training/training-line-fields-table-training/defining-tables-and-columns.md).

**AI Table Extraction:** Uses AI to extract the table of any supplier without training. If the results for one supplier are not accurate enough, train that supplier's table; the saved rules then take precedence over the AI for that supplier.

**Use Table Extraction Vision (AI):** Choose the input for AI table extraction from a four-mode dropdown. This is a mode selection, rather than a separate on/off switch. Changing the selected mode saves the setting immediately:

* **Off (text only)** – Only OCR text is used.
* **Images only** – Only the page image is read.
* **Hybrid (text and images)** – Text layer and page image are combined in the same request.
* **Auto (decided per document)** – DocBits chooses text only or hybrid input for each document.

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-556c539b84a3a84eecfebe16d318c1a914910357%2Fdbdc-3-four-vision-options-en-20261031.png?alt=media" alt="Current English Vision dropdown showing Off (text only), Images only, Hybrid (text and images) and Auto (decided per document)"><figcaption><p>The four input modes in the current English Dev menu. The menu was opened and closed without selecting a different mode; no extraction was run.</p></figcaption></figure>

**Use Structured Extraction (AI):** The switch sets the organization's structured-output default. The requested columns depend on the three-state Use AI selection; template overrides and supplier training also matter. Read the complete [configuration, preview and example guide](https://docs.docbits.com/end-user-and-partner-section/ai-table/structured-output.md) for Structured Output for Tables.

**Table Extraction for Costing Element:** When enabled, DocBits can extract costing elements from tables at the line level and classify them accordingly.\
Detailed explanation available [here](https://docs.docbits.com/administration-and-setup/settings/document-processing/classification-and-extraction/table-extraction-for-costing-element.md).

**Auto Extract Tax Code:** When enabled, the system automatically fills the **Tax Code** field on the Validation Screen, provided that a tax code field is configured.\
More information on this setting [here](https://docs.docbits.com/administration-and-setup/settings/document-processing/classification-and-extraction/auto-extract-tax-code.md).

**Save extraction rules (Admin only):** Only administrators can click *Save Rules* in table training. Switch it on when users keep saving rules that break a supplier's extraction.

**Extract hidden fields:** Fields marked as hidden in the field settings are extracted as well. Use it when a value is needed for a rule or an export but should not be shown on the validation screen.

**AI Model:** Sets the organization model for field and table extraction. The menu offers Full, Fast, Nexus, Nexus Flash, Auto and Auto Nexus. Check availability, quality, runtime and cost with representative documents. A tier is not always the most accurate or fastest for every document. Template overrides can replace the organization setting. The table below the selector shows:

* Which **suppliers** use which AI model
* Whether they use structured extraction (AI) and E-Text
* Options to delete an entry and thereby reset the supplier to the default

This setting is explained in detail [here](https://docs.docbits.com/administration-and-setup/settings/document-processing/classification-and-extraction/ai-model.md).

#### Why does the table look different per supplier?

Table processing depends on saved rules and settings in the relevant supplier or template context:

* **Saved rules** (table training): position of the table and mapping of its columns on that supplier's layout.
* **AI table tags and formatting rules**: hints the user saved for that supplier's AI table.
* **Template-specific AI model:** The selection made on the validation screen is saved per organization and current extraction template. It does not automatically change every template of the same supplier. Without an override, the applicable organization setting is used.

So supplier A with saved rules shows a deterministic table in the *Extracted table* tab of the validation screen, while supplier B without rules gets the *AI Extracted table*. To make supplier B behave like A, train B's table once. To reset a supplier, delete its rules on the validation screen or reset its training data in the AI Model table.

#### Preference keys

Every toggle in this section is stored as an organization preference. Use the key when you set the value through the API (`/preferences/set_preference`), a script, or the DocBits MCP (`get_preference` / `set_preference`).

| Setting (UI label)                                 | Preference key                   | Values                                                                                                                                  |
| -------------------------------------------------- | -------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------- |
| Split                                              | `SPLIT_DOCUMENT_ON_BARCODE`      | `true` / `false`                                                                                                                        |
| Barcode Type                                       | `SPLIT_DOCUMENT_BARCODE_TYPE`    | list of barcode type names                                                                                                              |
| Barcode Pattern                                    | `SPLIT_DOCUMENT_BARCODE_PATTERN` | regex string                                                                                                                            |
| Pages to process                                   | `RESTRICT_TO_PAGES`              | number; empty or `0` = all pages                                                                                                        |
| Allow Rounding During Amount Comparison            | `AMOUNTS_ROUNDING`               | `true` / `false`                                                                                                                        |
| Require Exact Match for Amount Comparison          | `AMOUNTS_NO_ROUNDING`            | `true` / `false`                                                                                                                        |
| Table Extraction                                   | `TABLE_EXTRACTION_SETTING`       | `true` / `false`                                                                                                                        |
| AI Table extraction                                | `USE_AI_TABLE_EXTRACTION`        | `true` / `false`                                                                                                                        |
| Use Table Extraction Vision (AI)                   | `TABLE_EXTRACTION_VISION_MODE`   | `off`, `vision`, `hybrid`, `auto`; when unset, the legacy preference determines the mode                                                |
| Use Table Extraction Vision (AI) – legacy switch   | `TABLE_EXTRACTION_USE_VISION`    | `true` / `false`; effectively means `vision`                                                                                            |
| Use Structured Extraction (AI)                     | `USE_STRUCTURED_EXTRACTION`      | `true` / `false`                                                                                                                        |
| Table extraction for costing element               | `CHARGES_TABLE_EXTRACTION`       | `true` / `false`                                                                                                                        |
| Auto extract tax code                              | `AUTO_EXTRACT_TAX_CODE`          | `true` / `false`                                                                                                                        |
| Save extraction rules (Admin only)                 | `ONLY_ADMIN_CAN_SAVE_RULES`      | `true` / `false`                                                                                                                        |
| Extract hidden fields                              | `EXTRACT_HIDDEN_FIELDS`          | `true` / `false`                                                                                                                        |
| AI Model                                           | `AI_MODEL`                       | `gpt-6-sol` (Full), `gpt-6-luna` (Fast), `qwen3.8-max` (Nexus), `glm-5.3-flash` (Nexus Flash), `auto` (Auto), `auto-nexus` (Auto Nexus) |
| Table extraction version (confirmation dialog)     | `TBL_EXT_VERSION`                | version string                                                                                                                          |
| OCR Settings → Use AI data for tables if available | `USE_AI_DATA_FOR_TABLE`          | `true` / `false`                                                                                                                        |
| OCR Settings → Use E-Text if available             | `USE_ETEXT_IF_AVAILABLE`         | `true` / `false`                                                                                                                        |

Notes:

* Boolean preferences are stored as the strings `true` / `false`. An unset key uses the default for that setting; the UI shows **Extract hidden fields** as on when unset. Do not assume every absent key means false. If you send `1` or `0`, DocBits stores `true` / `false`.
* `AI_MODEL` unset means **Fast**.
* `TABLE_EXTRACTION_VISION_MODE` unset is resolved from the legacy `TABLE_EXTRACTION_USE_VISION` switch: `true` resolves to `vision`, anything else to `off`.
* Changing a key takes effect for documents processed afterwards. Restart a document to re-extract it with the new setting.
* Template overrides differ from organization preferences. The AI Model guide explains the exact scope.

### Electronic Document

**Process Unsupported ZUGFeRD PDF:** If enabled, unsupported **ZUGFeRD** versions will be processed as standard PDFs, and the embedded XML will be ignored.

The list of supported **ZUGFeRD** versions can be found [here](https://docs.docbits.com/administration-and-setup/settings/global-settings/document-types/edi/zugferd.md).

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-aa8ed1cf779886825fc47d82b6303caa112aaa5c%2Fdbdc-193-electronic-document-en-20261030.png?alt=media" alt="Electronic Document with the Process Unsupported ZUGFeRD PDF switch"><figcaption><p>Electronic Document contains the option for unsupported ZUGFeRD PDFs. Its displayed state is an existing setting; no document processing was performed for this capture.</p></figcaption></figure>

## Classification Rules

The **Classification Rules** tab lists the rules that assign a document type automatically during processing. Each rule matches a **regex pattern** against the source name of the document and, on a match, sets the document type and optionally a sub document type.

Switch to the tab by clicking **Classification Rules** at the top of the page.

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-06af405be5fca2f5bd83670c15894dfe869da070%2Fdbdc-193-rules-list-en-20261030.png?alt=media" alt="Classification Rules with Search by column, Add and the rule-table headings"><figcaption><p>Search existing rules or choose Add to prepare a rule. Existing rule rows are obscured; this is not an empty-list example.</p></figcaption></figure>

The table has one row per rule with these columns:

* **Type** – where the pattern is applied. Two types exist: **Barcode** and **E-Mail**.
* **Pattern** – the regex stored for the rule.
* **Document Type** – the document type assigned when the pattern matches.
* **Sub Document Type** – the sub document type assigned when the pattern matches.
* **Sub Organization** – the sub organization the rule is limited to. Empty means the rule applies to all sub organizations.
* **Actions** – the row menu with *Edit* and *Delete*.

The search field above the table filters all columns, including the display labels of the dropdown columns.

### Add a New Classification Rule

1. Click **Add** above the table.

   <figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-eac45208e7c274567452ceee8d3eec11e4aba0e6%2Fdbdc-193-rule-add-en-20261030.png?alt=media" alt="Blank Configure classification Rule dialog with Pattern, Show help, Type, organization, document type, Cancel and Save"><figcaption><p>Add opens this blank rule form. Check the pattern and target type before Save; Cancel closes it without creating a rule.</p></figcaption></figure>
2. Fill in the fields:
   * **Pattern** (required) – the regex the rule looks for.
   * **Type** (required) – **Barcode** or **E-Mail**. This decides which source name the pattern is matched against.
   * **Sub Organization** *(optional)* – limits the rule to one sub organization.
   * **Document Type** – the document type assigned when the pattern matches.
   * **Sub Document Type** *(optional)* – a more specific type; the list depends on the selected document type. Changing the document type clears the sub document type.
3. Click **Save**.

The dialog validates the pattern before saving. A pattern that is already used by another rule is rejected with *Pattern already exists*; an empty pattern is rejected with the required-field message.

#### Pattern help

The **Show help** switch below the pattern field opens an inline reference with the source-name formats the pattern is matched against:

* **O365** – `Shared Email + Folder › Shared Email › Folder › Email › Id`
* **Gmail** – `Email + Folder › Folder › Email › Id`

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-c3da23f1a2c368cfa00ef35f0835b08400816d60%2Fdbdc-193-rule-pattern-help-en-20261030.png?alt=media" alt="Pattern help showing O365 and Gmail source-name formats, regular-expression examples and matching tips"><figcaption><p>Show help explains which source name the regex matches. The reference and examples are displayed without entering or saving a rule.</p></figcaption></figure>

Examples given in the panel:

* `.*@acme\.de.*` – any source from that mailbox
* `^ap@acme\.de:.*` – a specific mailbox, any folder

Tip: anchor with `^` / `$` for exact matches, escape dots as `\.`, and use `:` as the field separator. The first set of fields present in the source name wins.

If you use the **E-Mail** type, the pattern is matched against the mailbox and folder names of the imported mail, not against the document content.

### Edit a Classification Rule

1. Click the three dots in the **Actions** column of the rule you want to change.
2. Select **Edit**. The dialog opens with the rule's current values.
3. Make your changes to the pattern, type, sub organization, document type or sub document type.
4. Click **Save**.

The same validation applies as when adding a rule: the pattern must not be empty and must not duplicate another rule's pattern.

### Delete a Classification Rule

1. Click the three dots in the **Actions** column.
2. Select **Delete**.

The rule is removed from the list immediately. Deleting a rule does not change documents that were already classified with it.

## AI Models

The **AI Models** tab shows the custom-trained models available for document classification and their labels. It only appears when the model and label module is enabled for your organization.

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-55ab10f63c8455dc0249579d1a7ba93727f891ec%2Fdbdc-193-ai-models-empty-en-20261030.png?alt=media" alt="AI Models tab selected with No Record Found in the current organization"><figcaption><p>AI Models is an optional tab for custom classification models and labels. In this organization its successful list read returned no models; this does not describe the extraction-tier setting.</p></figcaption></figure>

If no custom model is available for the organization, the tab shows *No Record Found!*.

### Models

Each model is an expandable entry with:

* The **model name** and a badge **Default** on the model that is used when nothing else applies.
* The number of **labels** trained for that model.
* Per model, once expanded: **Type**, **First page only** and **Version**.

Below that information, the labels of the model are listed.

### Labels

Labels are what the model learns. Each label maps a pattern the model recognizes to a document type.

* **Name** – the label's name.
* **Next model** – a model that continues processing the output of this label.
* **DOC TYPE** – the document type assigned by the label.
* **Sub Document Type** – a more specific type for the label.
* **Priority** – the order in which the labels are evaluated; lower numbers first.

The search field above the label table filters the labels of the selected model.

### Editing a label

Click the pen icon on a label row to open the edit dialog. It offers **Next model** and **Document Type**; set one of them, since the dialog requires either a next model or a document type. Click **Save** to apply the change.

## Related pages

* [AI Model](https://docs.docbits.com/administration-and-setup/settings/document-processing/classification-and-extraction/ai-model.md) – the six extraction tiers and how a supplier gets its own model
* [Auto Extract Tax Code](https://docs.docbits.com/administration-and-setup/settings/document-processing/classification-and-extraction/auto-extract-tax-code.md)
* [Table Extraction for Costing Element](https://docs.docbits.com/administration-and-setup/settings/document-processing/classification-and-extraction/table-extraction-for-costing-element.md)
* [OCR Settings](https://docs.docbits.com/administration-and-setup/settings/document-processing/ocr-settings.md)

<figure><img src="https://578966019-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FT2n2w4uDCJvv7CJ5zrdk%2Fuploads%2Fgit-blob-879a814669b34cb0548370bb180ae643882e11b6%2Fstructured-output-settings-en-20261104.png?alt=media" alt="Current Extraction controls. Structured Output is off in this existing configuration, so no schema preview is shown. No extraction setting was changed."><figcaption><p>Current Extraction controls. Structured Output is off in this existing configuration, so no schema preview is shown. No extraction setting was changed.</p></figcaption></figure>


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.docbits.com/administration-and-setup/settings/document-processing/classification-and-extraction.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
