Revise launches an AI document editor with inline rewriting, revision history, and collaboration

• Revise launched as a document-first AI editor that keeps proofreading, rewriting, editing, and generation inside the page rather than sending writers out to a chatbot, according to the launch thread.
• The product page says Revise also ships with visual revision history, real-time collaboration, and import/export for Word, PDF, Markdown, and HTML via the product page.
• Revise supports multiple model providers — GPT-5.4, Claude, and Grok — while aiming at essays, reports, blog posts, and other longform writing workflows, per the feature list.
• Early discussion on Hacker News focused less on generic AI writing and more on whether the inline workflow can help with clarity, structure, and reordering arguments, as the discussion summary shows.

Revise is positioning itself as a full word processor with an embedded AI agent, not a chat sidebar bolted onto docs. On the product page, the feature set includes inline rewriting and proofreading, revision history, collaboration, and richer document elements such as tables, code, equations, diagrams, and images. The same page says it can import and export .docx, PDF, Markdown, and HTML, which makes it more relevant to writers moving drafts between publishing and client workflows.

The launch thread frames the main pitch more bluntly: the founder's post says the goal is to replace the copy-paste loop between ChatGPT and a document when you are iterating heavily on a draft.

The most useful early signal is what people wanted the tool to do. In the discussion, one commenter asked for help making writing and thinking clearer — checking whether a sentence makes sense, whether a conclusion follows, and whether an essay could be rearranged to read more simply. That is a stronger test than one-off text generation because it centers structure and reasoning inside the draft itself.

There is also some technical ambition under the hood. The founder said in the launch thread that Revise was built from scratch, using canvas-based text rendering and a custom block layout engine for lists and paragraphs, which suggests the editing experience is meant to be a core product surface rather than a thin wrapper around model calls.