{"slug": "versionstory-comparison-and-redlining-for-ai-native-legal-teams", "title": "VersionStory: Comparison and Redlining for AI native legal teams", "summary": "VersionStory, a Rust-based document comparison and redlining engine for legal teams, reported that in a September 2026 test of 300 Claude Opus 5 agents summarizing the same 100 production redlines, Markdown output was faster than Word in 99 of 100 cases and faster than PDF in all 100, typically requiring one tool call versus four for Word and nine for PDF. The engine accepts .DOCX, legacy .DOC, native .PDF and scanned .PDF via OCR, and exports .DOCX, .PDF, changed-pages-only .PDF, .MD and .JSON, all derived from a single comparison. VersionStory states its AWS architecture and operational controls are independently examined under SOC 2 Type 2, that uploaded documents are never used to train a model, and that third-party penetration test reports are available under NDA.", "body_md": "## Say “redline this” in your AI.\n\n## Any file in. Any format out.\n\nThe Rust-based engine reads whatever lawyers send and returns the redline in the format you ask for. Lawyers get Word or PDF. Agents get Markdown. Your code gets JSON.\n\nInput\n\n- .DOCX OpenXML standard\n- .DOC Binary format (legacy)\n- .PDF(NATIVE) Digital PDF\n- .PDF(SCAN) Scanned pages, read with OCR\n\nOutput\n\n- .DOCX For lawyers and agents to edit\n- .PDF For lawyers to review\n- .PDF(CHANGED PAGES) For efficient lawyer review\n- .MD For quick, token-efficient agent review\n- .JSON For parsing in your deterministic apps\n\nOne comparison in the middle · every output derived from the same diff\n\n## Agents read Markdown redlines\n\nin one step.\n\nWe handed 300 agents the same 100 production redlines as Markdown, Word, and PDF and asked each to summarize the changes. Markdown was faster than Word in 99 of the 100 and faster than PDF in all 100, and it typically took one tool call where Word took four and PDF took nine.\n\n### Time to summarize a redline\n\n- VS Markdown\n- .docx\n\naverage seconds per redline\n\n0s20s40s60s80s\nUnder 25 KB\n\n25–75 KB\n\nOver 75 KB\n\n### AI tokens to summarize a redline\n\n- VS Markdown\n- .docx\n\naverage tokens per redline\n\n0200k400k600k800k\nUnder 25 KB\n\n25–75 KB\n\nOver 75 KB\n\nMethod: 100 comparisons drawn at random from redlines created through the public API, each delivered in all three formats from the same comparison. One independent Claude Opus 5 agent per redline per format, given the same instruction. Groups are by Markdown file size; values are means within each group. Run September 2026.\n\n## Precision in\n\nEvery Element\n\nVersion Story redlines Word and PDF to the highest standards of precision — down to the footnote, the cross-reference, and the scanned page.\n\n### Upload Formats\n\n- Word docs to PDF docs\n- PDF docs to PDF docs\n- Word docs to Word docs\n- Scanned PDF docs (OCR capabilities)\n\n### Export Formats\n\n- PDF with changed-pages-only\n- PDF with the full document\n- Word with tracked changes\n- … with a changes summary page at the end\n\n### Paragraphs\n\n- Changes to words and characters within a paragraph\n- Insertions or deletions of whole paragraphs\n- Numbering changes\n\n### Tables\n\n- Changes to text within a table cell\n- Insertions or deletions of cells, rows, or columns\n- Numbering changes\n\n### Section Cross References\n\n- Changes to the number or letter of the referenced provision\n- Linking preserved\n- Formatting preserved\n\n### Table of Contents\n\n- Changes to name, number, or page number of sections\n- Linking preserved\n- Formatting preserved\n\n### Comments\n\n- Changes to text within a comment\n- Insertions or deletions of whole comments or threads\n\n### Headers & Footers\n\n- Changes to words and characters\n- Insertion or deletion of elements (e.g., page numbers)\n\n### Footnotes & Endnotes\n\n- Changes to text within a footnote or endnote\n- Insertion or deletion of whole footnotes or endnotes\n\n## Security your risk committee will sign off on.\n\nLaw firms hand us their most sensitive drafts. The controls are the same whether the request comes from a lawyer, an API key, or an agent.\n\n- ### Your data, your controlDocuments stay in your workspace until you delete them, admins can set automatic deletion for inactive projects, and nothing you upload is ever used to train a model.\n- ### SOC 2 Type 2Our AWS architecture and operational controls have been independently examined under SOC 2 Type 2, so our security controls are verified by a third party, not asserted by us.\n- ### Independent penetration testingThird-party penetration testing, alongside an A+ Security Scorecard rating and continuous monitoring, with the reports available under NDA.\n\n## Loved by lawyers\n\n## Deterministic tools\n\nfor every file format\n\n### Compare\n\nOur rules-based redline engine, written in Rust, guarantees the precision you need to harness LLM based workflows, and scales to thousands of .docx, .doc, and .pdf files.\n\n### Track\n\nEvery draft of a document, one history. Track layers each turn of a negotiation into a single tracked-changes document, attributing every surviving change to the version and author that introduced it. Blame for contracts, built the same deterministic way.\n\n### Merge\n\nThree reviewers, three marked-up copies, one draft. Merge reconciles parallel revisions of the same base into a single tracked-changes document, attributes each edit to its source, and flags the places where reviewers overwrote each other instead of silently picking a winner.", "url": "https://wpnews.pro/news/versionstory-comparison-and-redlining-for-ai-native-legal-teams", "canonical_source": "https://www.versionstory.com", "published_at": "2026-10-04 04:55:00+00:00", "updated_at": "2026-10-04 05:07:37.899229+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "ai-products", "developer-tools"], "entities": ["VersionStory", "Claude Opus 5", "Anthropic", "AWS", "SOC 2 Type 2"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/versionstory-comparison-and-redlining-for-ai-native-legal-teams", "markdown": "https://wpnews.pro/news/versionstory-comparison-and-redlining-for-ai-native-legal-teams.md", "text": "https://wpnews.pro/news/versionstory-comparison-and-redlining-for-ai-native-legal-teams.txt", "jsonld": "https://wpnews.pro/news/versionstory-comparison-and-redlining-for-ai-native-legal-teams.jsonld"}}