What is markform?

Markform is a TypeScript library that adds structured, fillable fields to plain Markdown documents. The idea is that a single .form.md file can serve as both a human-readable document and a precise data collection form that AI agents or users can fill in step by step. You write a Markform document like any Markdown file, but embed field definitions inside HTML comment tags. Because HTML comments are invisible when rendered on GitHub or in a browser, the document looks like normal Markdown to a reader. Under the hood, each comment tag declares a field with a type (text, number, checkbox, table, dropdown, and so on), a label, validation rules like minimum length or required status, and a role indicating whether a human or an AI agent is expected to fill it. When an agent fills in a form, it submits individual patch operations targeting specific fields by their ID. Each patch is validated against the rules defined in that field, so errors are caught immediately and the agent can correct them without resubmitting the entire form. Large forms with hundreds of fields can be filled by many concurrent AI requests, which makes the format useful for large research workflows where many questions need answers in parallel. The format can be exported to plain Markdown (for human reading), JSON or YAML (for downstream processing), or JSON Schema (for use with other tools). A command-line tool called markform serve starts a small local web server for viewing and debugging forms, including a visual overview of which fields have been filled and how long each one took. The project includes integration with the Vercel AI SDK, so agents built on that framework can call the form-filling tools directly. Installation is via npm. The full README is longer than what was shown.