Release Notes

What's new in each version of MDraft.

1.0 — initial release (2026)

A native macOS workspace for AI prompts, instructions, and Markdown-based workflows. Preview / Markdown / Split editing, discovered AI Context Files (AGENTS.md, CLAUDE.md, GEMINI.md, SKILL.md, Cursor rules, Copilot instructions, Windsurf rules), a Library for capturing and re-deploying reusable AI documents, sidebar keyword search across open docs / Library / discovered files / your added folders, local Semantic Search across every indexed document, Context Bundle export, and one-click Web Page import that pulls articles into editable Markdown with image sidecar folders.

Highlights

• Editor. Preview, Markdown, and Split modes. WYSIWYG editing with full Markdown round-trip. Tables, task lists, inline images, syntax-highlighted code blocks, blockquotes, thematic breaks. Lossless save / reload.
• Sidebar. Four sections — Open Editors, Folders, AI Context, Library — plus a keyword search bar that searches names and content across all four sources at once.
• AI Context Discovery. Automatic scan of your home directory for AI-tool configuration files; live-updated via FSEvents on external file changes.
• Library. Capture markdown documents into a local SQLite store, classify by kind, deploy back to disk with sensible default filenames. Optional Foundation-Models-assisted classification on macOS 26+ with Apple Intelligence enabled.
• Web Import. Drop a URL into the toolbar's Import panel, get a saved Markdown document with Readability-extracted content, downloaded images in a sibling assets folder, and mdraft.source.* provenance frontmatter. Refresh from Source re-fetches and diffs at chunk level, preserving your local edits.
• Semantic Search. Local NLEmbedding-based similarity search across every saved document. No cloud, no network, no third-party SDKs.
• On-Device AI. On Apple-Intelligence Macs: selection summarization, document summarization, metadata extraction, token-budget compression, tag generation, and instruction-file validation. Foundation Models, on-device, no upload.
• Context Bundle Export. Deterministic, byte-stable merged markdown for LLM context injection.
• Templates. Built-in scaffolds for AGENTS.md, CLAUDE.md, SKILL.md, prompt specs, evaluation docs, research briefs, context-pack indexes.

Requirements

macOS 14 (Sonoma) or later. Universal binary; runs natively on Apple Silicon and Intel. Apple Intelligence features additionally require macOS 26 (Tahoe) on supported hardware.

Privacy

No analytics, no telemetry, no third-party SDKs, no data collection of any kind. The only network calls are the ones you trigger explicitly: fetching a web page for import, loading a remote image referenced from a markdown document, and the system's own one-time download of the NLEmbedding language model the first time you use Semantic Search. Everything else is local.