Documentation

Getting started

Ristmark turns your existing Word and RTF documents into reusable, data-driven templates — then generates clean PDF, HTML or DOCX from them, on demand or via API.

1. Migrate a document

Click Migrate a document (top of the sidebar) and upload a .docx or .rtf file — PDF and image files are accepted too. Only the first 7 pages are migrated; a longer document is truncated there and Ristmark tells you it happened. Ristmark renders each page and rebuilds its layout from the picture (every migration is a visual reconstruction, in Beta), then labels the parts that vary — names, dates, amounts — as fields. Every field gets a confidence score, so the ones the AI is unsure about flag themselves. Word and RTF also provide a separate structural cross-check. PDF and image reconstruction depends on the visible page content without that check. Review the result against your original for every file type.

2. Review the result

You land in the review view: your reconstructed document with its fields marked by confidence band — Confident, Worth a glance or Not sure — check this. Use Next check to work through field decisions, fixed values and layout checks. The selected check has visible actions: Looks right, Fix this or Keep fixed. A confidence score describes the model’s opinion; compare the actual values and output with your original, including amounts and dates. Your corrections are remembered — the next similar document applies them automatically. When you’re done, click Approve: the review closes, your verdicts are written into the template, and it moves from Review into Templates. Anything you didn’t check stays as proposed; you can still change it in the editor.

3. Edit the template

Open the template in the editor. It works like a familiar document editor: double-click a block or field to edit it, change text, adjust layout, add or remove blocks. Fields you mark as data-bound pull from your data at generation time; everything else is fixed boilerplate.

4. Generate

Render your template to PDF, HTML or DOCX — from the app (open Preview, pick the format, then Download final), or programmatically via the /v1 API. No AI runs at generation time, so the same data produces the same PDF and HTML every time. (DOCX is an editable Word file and is not yet byte-stable across renders.)

PDF is the pixel-faithful format. DOCX is a genuinely editable Word document rather than a copy of the canvas, and has known limits: a letterhead background is not carried into DOCX yet — only the PDF render includes it; table borders are applied per table rather than per cell; lists (flat or nested) export as ordinary paragraphs with their bullet or number characters written into the text; and a ruled form keeps its text but not its drawn layout — labels and values print one under another. The app shows the same note next to the DOCX preview.

Have a batch of records instead of one? See Generate from a spreadsheet — upload a CSV or Excel file and generate one document per row.

5. Consolidate near-duplicates (optional, Beta)

Consolidation is in Beta. Its matching is text-based and has not been validated for accuracy. Open Similar to see suggestions based on shared wording. Similar wording does not prove that documents have the same meaning or layout.

Choose Review matches, select the documents to include and choose a base. Review differences shows source values, proposed fields, layout differences and content that would not be included. Simple value changes can become fields; changed clauses or unsupported differences require manual consolidation using a copy of the base.

After checking the plan, choose Create review draft. The draft opens in Review for approval. The generation API requires an approved template; you can preview the draft while reviewing it. Your originals stay unchanged. Return to Similar to resume a pending review or open an existing result; archive originals from Templates only when you are satisfied with the approved replacement.