Tags and extraction

Tags define what TextMine should extract, review, and reuse from Vault documents. They can be document-type defaults, one-off document questions, reusable custom tags, chained outputs, entity extraction prompts, image tags, or table-aware questions. Use this page when designing extraction configuration or explaining how Vault tags work.

​

Tag scopes

A tag result can include a value, reasoning, source passages, preview offsets, page number, validation state, status, model metadata, and whether the tag came from a default configuration.

​

Default tags

Default tags are reusable extraction fields for a document type. They are the best fit when the same field should be extracted every time a document of that type is processed. Use default tags for stable fields such as:

• Effective date.
• Counterparty.
• Governing law.
• Termination notice period.
• Contract value.
• Renewal date.
• Supplier name.
• Clause presence or absence.

Default tags can be copied, reused, and applied when documents are processed or reprocessed. If a document type or its default tags change, affected documents may need reprocessing before results reflect the new configuration.

​

Custom tags

Custom tags are user-defined questions that can be run against selected documents or a Vault scope. They are useful when the field is not part of the standard document type or when a team needs temporary or evolving extraction. Common custom tag modes include: Custom tags can also request reasoning, produce semantic-card style output, or be chained from earlier extraction context, depending on configuration.

​

Aggregation tags

Aggregation tags use a deeper retrieval and reasoning path. Use them when a simple field extraction is too narrow. Good examples:

• Summarise all payment obligations in this agreement.
• Identify every termination right and group by party.
• Compare all references to change-of-control restrictions.
• Aggregate all dates that affect renewal, notice, or expiry.

Avoid aggregation tags when the desired answer is a single obvious field. A default or fact-style tag is usually easier to validate and rerun.

​

Chained tags

Chained tags use earlier context as inputs for a new question. A chained tag can reference:

• Other custom tags.
• Default tags.
• Extracted entities.
• Extracted tables.

Use chained tags when the answer depends on intermediate extraction. For example:

• Use extracted parties and governing law to assess jurisdiction-specific risk.
• Use table extraction as context for a covenant or pricing answer.
• Use entity extraction plus contract dates to build a party-obligation summary.
• Use a default tag value as a filter or comparison point for a custom tag.

Chained tags cannot be treated like independent extraction. Agents should inspect the source tags first, confirm their status, and rerun upstream tags or tables if the context is stale.

​

Entity tags

Entity tags use entity resolution. Instead of a free-form question, the user defines entity names or categories, optionally with descriptions. TextMine extracts and resolves those entities from the document context. Use entity tags for:

• Parties and affiliates.
• People, roles, signatories, or contacts.
• Products, assets, facilities, suppliers, or properties.
• Legal concepts that should become nodes in a graph.
• Domain-specific entities that Records or Workbench should use later.

Entity extraction can feed the Vault entity graph, Records, workflows, and chained tags. Include descriptions when entity names are ambiguous.

​

Image tags

Image tags use multimodal question answering over an image region. They can be created from a selected screenshot region in the Vault editor or from embedded image detection. Image tags store image context such as page number, coordinates, width, height, and the uploaded image asset. They are useful for:

• Ownership charts.
• Signature blocks.
• Scanned tables.
• Organograms and diagrams.
• Embedded images that contain legally or commercially meaningful text.

Image tags are different from text tags. Aggregation and semantic-card modes are disabled for image tags because the source is the selected image context rather than text retrieval over the full document.

​

Embedded image detection

Embedded image detection is a Vault/document-type processing setting. When enabled, TextMine can render document pages, classify visual regions against configured image types, and apply image-type questions to detected image content. The flow is:

1. Configure image types and image tags for the Vault or document type.
2. Enable embedded image classification/image processing where appropriate.
3. Process or reprocess documents.
4. TextMine detects matching embedded image regions.
5. Matching image tags run as multimodal questions against the detected image context.
6. Users review the image-tag answers in the editor like other extracted values.

Use embedded image detection when important evidence is visually embedded rather than available as ordinary text. It is especially useful for charts, scanned excerpts, diagrams, and visual schedules.

​

Table extraction

Table extraction detects tables and line items from documents so they can be reviewed, exported, used by reports, or used as context for chained tags and workflows. Use table extraction for:

• Pricing schedules.
• Fee tables.
• Covenant calculations.
• Line-item invoices.
• Obligation matrices.
• Disclosure schedules.
• SEC filing tables.

Tables can be viewed in the Vault editor and in a pop-out table view. Workbench and workflows can also queue table extraction and wait for completion before using table outputs.

​

Extraction settings

Vault/document-type settings can control several processing behaviors: These settings may be inherited from Vault defaults or overridden at a document-type/Vault association. Reprocess documents after changing settings if existing results need to be refreshed.

​

Status and validation

Tag answers move through processing states and may need human validation. A good review should check:

• The extracted value.
• The source passage, table, entity, or image region.
• The page and preview offsets where available.
• Whether the tag is current with the latest document type/settings.
• Whether upstream chained sources completed successfully.

For durable downstream use, prefer validated tags or outputs with clear source evidence.

​

Agent guidance

Agents should not create tags blindly. Before creating or running tags, they should:

1. Read existing document types and tags.
2. Prefer default tags for stable document-type fields.
3. Use custom tags for team-specific or evolving questions.
4. Use aggregation tags only when synthesis is needed.
5. Use chained tags only after checking upstream tags, entities, or tables.
6. Use entity tags when output should feed graphs, records, or relationship-aware workflows.
7. Use image tags for visual evidence.
8. Ask for approval before creating tags, running tags across many documents, changing extraction settings, or reprocessing documents.