Skip to content
VeraFrame Docs

Document templates

Many teams already have Word or Excel templates for the documents they produce — quotes, comparison reports, proposals, contract drafts, compliance summaries. Document templates let you use those same files as the output format for VeraFrame.

You upload the template. VeraFrame scans it for placeholders. At generation time, VeraFrame fills each placeholder with content that has been verified against your source material, and returns the rendered .docx or .xlsx file.

Document templates require the document_templates feature.

The workflow

  1. Upload the template. Mark up the fields you want filled with {{placeholder_name}} syntax in the Word or Excel document.
  2. Scan the template. VeraFrame detects all placeholders and builds an initial schema.
  3. Define the schema. For each placeholder you set a description, an expected data type, and — for fields that must come verbatim from a source document — a placeholder class.
  4. Activate the template. It becomes selectable in the Generate mode.
  5. Generate. Select the template in a Generate request. The AI produces placeholder values from your source material. The validator checks each value. The rendered document is returned together with the trust report.

Placeholder classes

Each placeholder carries a class that tells the retrieval and validation pipeline how strict to be.

  • fact placeholders — the normal mode. The AI composes the field from source material and the value is then checked against the selected evidence.
  • whole_block_text placeholders — the field must come from a single, complete block of source text, used verbatim. Use this for legal clauses, SLAs, contractual terms, approved boilerplate — anywhere the exact wording matters and paraphrasing is unacceptable.

The whole_block_text class is the main reason document templates exist in regulated workflows: it rules out the possibility of the AI silently rewording a contract clause.

The block-based trust pipeline

Behind the scenes, every placeholder is filled through a three-step process:

  1. Retrieve. VeraFrame runs a full-text search over the source group and pulls a small set of candidate blocks.
  2. Score. Each candidate is scored for relevance to the placeholder’s description, its expected type, and any numeric constraints. The scores are surfaced in a retrieval audit.
  3. Select and verify. The top-scoring block (or the top whole_block_text block verbatim) is used to fill the placeholder. The validator then checks the filled document against the same blocks — so the generator and verifier agree on the source.

If the top candidates are too close to one another, the placeholder is flagged needs_review in the trust report.

Supported template formats

  • Word (.docx) — paragraphs, tables, headers, and footers are scanned.
  • Excel (.xlsx) — cell-level placeholders are scanned and rendered.

An optional metadata table inside the template can define placeholder descriptions and types without going through the UI — useful when your templates are authored by subject-matter experts who prefer to stay in Word or Excel.

Current limitation:

  • Excel rendering supports single-cell placeholder filling only. Multi-row or table-style output is intentionally not rendered into .xlsx.

Next steps