marka.md review: a local-first markdown editor built around AI context bundles

The problem marka.md targets: assembling context for an AI prompt

Most markdown editors assume the document is the destination. marka.md assumes the document is an input to something else. The README frames the whole application around a single loop, quoted as "collect notes → write → share with ai", and states that nothing leaves the machine until you copy. That is the design constraint everything else follows from.

The practical problem is mundane and familiar. You have a folder of notes. You want to hand four of them to a model along with a question. Doing that by hand means opening each file, copying its contents, pasting, and then manually typing the filenames so the model knows which text came from where. marka.md puts a staging area, called a context tray, next to the sidebar. You stage files there, the app shows file and token counts, and one command copies an AI-ready bundle with relative paths preserved.

Who this is for: someone who already writes markdown locally, already uses a chat model or a local agent, and finds the copy-paste assembly step annoying enough to want tooling for it. The README says the output works with Claude, ChatGPT, Gemini, local agents, and anything that reads plain markdown, which is a deliberately low bar. It is not a note-taking system with a database, and it is not a sync service.

Architecture: Tauri 2.11 shell, React 19 frontend, Codemirror 6 editor

The stack table in the README is unusually explicit, and the repository layout matches it. src-tauri/ holds the Rust side. src/ holds the frontend. The package.json confirms the split: @tauri-apps/api, plugin-dialog, plugin-fs, plugin-opener, plugin-process and plugin-updater are runtime dependencies, while the editor is codemirror plus @codemirror/lang-markdown and @codemirror/search. Markdown rendering goes through markdown-it with markdown-it-mark and markdown-it-task-lists, syntax highlighting through shiki, and diagrams through mermaid. Both shiki themes and mermaid are loaded lazily, which the README lists as a v1.6.1 change for markdown highlighting.

The important consequence of the Tauri choice is that file access runs through Rust plugins rather than a browser file API. That is what makes native file watching, folder monitoring, reveal-in-file-manager and undo of sidebar file operations possible at all. It also means the desktop build is not a wrapped web app with a filesystem shim; the window is a webview, but the file operations are native.

On the frontend side, styling is CSS variables with no framework, and i18next is a dependency, which lines up with the README's mention of interface languages and Traditional Chinese localization in v1.6.0. The editor is opt-in vim via @replit/codemirror-vim, so the default keymap stays conventional unless you enable it. TypeScript is pinned at ~7.0.2 and Vite at ^7.0.4, with engines requiring Node 24 or newer.

Installing marka.md and staging a first context bundle

The README points to the latest GitHub release as the primary download. On macOS, Homebrew is the shortest path, and the cask name is namespaced to the author's tap.

brew install --cask mattenarle10/tap/marka-md

After that, updates can go through Homebrew with the command below, or through the in-app signed updater, which the README says still works.

brew update && brew upgrade --cask marka-md

On Linux, the AppImage needs an explicit execute bit before it will run. The README gives no install step beyond that for the AppImage, and describes it as self-contained.

chmod +x marka.md_*.AppImage
./marka.md_*.AppImage

Debian and Ubuntu users get a .deb instead, installed with dpkg. Fedora and openSUSE users get a .rpm through dnf. Arch users can pull markamd-appimage from the AUR with yay or paru.

If you would rather build it, the README requires bun, rust, and platform build tools, plus libwebkit2gtk-4.1-dev and libsoup-3.0-dev on Linux. The two commands it gives are bun install and bun run tauri dev for a native window with hot module replacement. bun run tauri build produces the .dmg, .exe, .AppImage, .deb and .rpm under src-tauri/target/release/bundle/.

For a first real use, open a folder of notes with the folder shortcut, which the README lists as Command-Shift-O on macOS and Ctrl-Shift-O elsewhere. Stage the files you want from the sidebar into the context tray. The tray reports file and token counts. Then open the command palette with Command-K, or Ctrl-K, and run the copy context bundle command. The clipboard now holds one markdown document containing the staged notes with relative paths. Paste it into your model of choice. That is the entire loop, and the README does not document a preview step for the bundle before it reaches the clipboard.

Where marka.md falls short: bundle format, signing, and scope

The clearest gap is the bundle itself. The README states that the copied context includes relative paths and that file and token counts are visible in the tray, but it does not show a sample of the output. If your target model or agent expects a particular delimiter convention, XML tags, or a specific header per file, you cannot tell from the documentation whether marka.md produces it. The roadmap lists context handoff presets for bring-your-own-AI workflows, starting with markdown and XML-tag bundle formats, as unfinished work. Until that ships, the format is whatever the current build emits, and the README does not pin it down.

Signing is asymmetric across platforms. The README notes Apple notarization in the badge row, and the macOS updater is described as signed. The Windows build is explicitly unsigned, and the README warns that SmartScreen may ask for confirmation. That is a real friction point for anyone distributing the installer inside an organization with application control policies.

Scope is the third limitation, and it is a deliberate one. There is no plugin API in the dependency list or the README, no extension mechanism beyond themes and interface languages, and no mention of collaboration, comments, or a sync backend. If your team's notes live in a shared workspace with concurrent editing, marka.md is the wrong tool. It is a single-user local editor with a clipboard as its integration boundary, and the README treats that boundary as a feature rather than a missing piece.

How marka.md differs from Obsidian and VS Code with markdown extensions

Obsidian is the closest comparison in spirit, since both are local-first markdown tools where the files stay on disk. The difference is the unit of work. Obsidian organizes around a vault, backlinks, and a graph of notes, with plugins supplying almost everything else. marka.md organizes around a tray of staged files and a single copy action. Obsidian has a far larger extension surface and a mobile client; marka.md has neither, and the README does not claim otherwise. If you want a knowledge graph, Obsidian is the answer. If you want to hand six notes to a model in one paste, the tray is the point.

The other comparison is VS Code with a markdown preview extension. That gets you a capable editor, git integration, and a terminal in the same window, and it is already installed on many machines. What it does not give you is a staging area with token counts and a one-command bundle copy. You would assemble the context yourself, either by hand or with a script. marka.md trades the general-purpose editor environment for one narrow workflow done well.

There is a build-it-yourself option too, and it is worth naming because it is cheap: a shell loop that concatenates selected files with their paths into a temp file does most of what the tray does. marka.md's advantage over that is the visual selection, the token count, and the fact that the loop is one keystroke from the editor where the notes were written.

Maintenance, licensing, and what an upgrade actually costs

The repository is not archived. The last push was on 2026-09-14, and v1.7.4 was released the same day, with v1.7.3 on 2026-09-01 and v1.7.2 on 2026-08-15. That is a steady patch cadence over the preceding six weeks, and the roadmap shows completed items through v1.7.1 plus two open items: native or silent PDF generation, and context handoff presets. Release work follows a checklist in docs/release-checklist.md, and per-release detail lives on the project's changelog page.

Upgrade cost is low by design. The macOS Homebrew path is a single upgrade command, and the in-app updater is signed and described as still working alongside it. Windows and Linux users re-download installers, since the README does not describe a package manager path for Windows and only lists AUR for Arch on Linux. There is no documented migration step between versions, and no schema or database to migrate, because the application reads and writes plain markdown files you already own.

The licence is MIT, stated in the README badge row and in the LICENSE file at the repository root. That permits commercial use and modification, and it means the context tray's bundle format is not something you can demand changes to. The README links an Open Collective for funding. Nothing in the README suggests a dual licence, a contributor licence agreement, or a paid tier, but this is not legal advice and the LICENSE file is the authority.

Editorial conclusion

Adopt marka.md if you keep plain markdown notes locally and regularly paste them into Claude, ChatGPT, Gemini or a local agent, and you want the staging step inside the editor rather than in a scratch file. Skip it if you need real-time collaboration, a web client, or a plugin API, none of which the README describes. Before committing, verify two things on your own machine: that the Windows build triggers SmartScreen as documented, and that the context bundle it copies has the relative-path format your target model expects, since the README states the format but does not show a sample bundle.

Frequently asked questions

Official sources

1. License: MIT
2. mattenarle10/markamd on GitHub
3. Project website
4. README
5. Releases