Markra: a local-first WYSIWYG Markdown editor with AI edits you approve before they land
A WYSIWYG Markdown editor with native AI. Fully open source. Free to use. Your data stays local.
At a glance
- What is it?
- Markra is an AGPL-3.0 Markdown editor built on Tauri with a web companion, keeping plain .md files on disk and adding native AI commands, an edit preview step, multi-provider settings, WebDAV sync and local spellcheck without requiring an account.
- Who is it for?
- Take Markra if you write Markdown in folders on disk and want AI involved without handing your library to a cloud editor, because the file model is plain .md, the AI actions show a preview before anything is applied, providers are yours to configure, and the Tauri desktop build stays light. Prefer Typora if you only want minimal live preview, and Obsidian if the point is a linked knowledge base rather than document writing, which is the comparison the README itself draws.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository last received commits 3 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 17, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What Markra is and where it runs
Markra describes itself as a local-first, open-source Markdown editor that brings AI into the writing flow, and the structure matches the claim. Editing happens in WYSIWYG or source mode, everything stays as plain .md files on disk, and no account is required. The desktop app is built on Tauri and ships for macOS Apple Silicon and Intel, Windows as installer or portable, and Linux as AppImage, DEB, RPM and Arch packages through GitHub Releases. There is also a web editor at editor.markra.app with the full editing experience and browser file handles instead of native dialogs. On macOS the install is a Homebrew cask.
brew install --cask markrahq/tap/markraOn Arch Linux the release package installs through pacman.
sudo pacman -U ./Markra_<version>_linux_x64.pkg.tar.zstFiles stay on disk by default, and the only network contact is what you configure: optional WebDAV sync, remote image storage, and AI requests to the providers you set up.
The AI workflow: commands, preview, providers
The AI layer has three surfaces. An inline AI command bar handles polish, rewrite or expand actions on selected text. An AI edit preview step shows what the model wants to change before it lands, which the README calls full preview before any change, and this is the design decision that separates a writing tool from an autopilot. An AI side panel carries document context for longer conversations about the current file. Provider settings are multi-provider, so you bring your own model access, and the desktop app makes native runtime requests with app proxy settings while the web editor is limited to browser requests that respect provider CORS. The comparison table also names ACP local agents as part of the workflow, meaning the editor can reach locally running agents, not just hosted APIs. Nothing here routes your documents through the project servers, because there are no project servers in the path.
Desktop versus web in practice
The README carries a capability table that is honest about where the web editor falls short. Both surfaces give the full WYSIWYG and source editing. Desktop opens local files and folders with native dialogs, file paths and watchers, while web uses the browser picker and file handles. File tree operations shrink on the web to create, rename, move and delete where browser permissions allow, against full sort, reveal and multi-select on desktop. Auto-save and restore covers existing files, tabs, drafts and workspace windows on desktop, and browser state on web. Spellcheck is desktop only: Markra-managed local dictionaries with on-demand language packs and a personal dictionary. Image storage spans local folders, WebDAV, PicGo or PicList and S3-compatible storage on desktop, narrowing to local handles and WebDAV where CORS permits on web. Backup and note sync are desktop features, and export spans HTML, PDF and Pandoc formats when configured on desktop, against HTML download and browser print on web.
Where it sits against Typora and Obsidian
The README states plainly that Markra is not trying to replace every Markdown tool and calls itself closest to a calm document editor with native AI and local files. Against Typora, the difference is the AI workflow and the editing surface: Typora is minimal live preview with Markdown syntax hidden while writing and AI is not a core workflow, while Markra offers a WYSIWYG document surface plus source mode with native inline actions, a side panel and local agents. Against Obsidian, the difference is the product goal: Obsidian is a personal knowledge base with linked notes and a vault, while Markra keeps tabs, outline, workspace search and double-bracket link completion but does not build a graph. The practical read is that writers who want one document at a time with an AI editing pass fit Markra, researchers building connections fit Obsidian, and minimalists who resent both fits Typora.
Release history and what it says
The recent changelog reads like editor polish in the details that daily writing exposes. Version 2.10.3 on 2026-09-04 stabilized Windows frontmatter input and window sizing and kept IME input inside table headers, which matters for Chinese and Japanese typing. Version 2.11.0 on 2026-09-13 added the OrcaRouter provider and editable HTML tables with merged cells, applied the saved theme before window startup, and stopped ordered list markers from wrapping. Version 2.11.1 on 2026-09-17 switched to proportional quotes in non-Chinese locales, handled directory picker cancellation on web, made the AI wait for streamed chunks before closing responses, and scaled table text with the body font size. The workspace is a pnpm monorepo at version 2.11.1 with the desktop app under a Tauri script, and contributors MuRong and Zhucong appear in the notes, so development is active and outside contributions are landing.
Limits to weigh
The licence is AGPL-3.0, which is fine for personal use and a real consideration if you want to embed the editor or its components in a hosted service, since the network copyleft terms then apply to your product. The AI layer is only as good as the provider you configure, and the web editor strips the feature list down in ways the table makes clear, so web is a companion rather than an equal. Local spellcheck dictionaries download on demand, which is worth knowing on offline machines. The project has no sync server of its own: WebDAV or S3-compatible storage is the sync path, so teams used to first-party cloud sync need to bring infrastructure. And the AI edit preview, the feature that makes the workflow trustworthy, also means one extra confirmation step that fast typists may find slow until they trust a given prompt.
Who should pick it
The audience is writers and developers who keep their notes in Markdown files, want an editing surface nicer than a code editor, and want AI available on tap without giving up file ownership. Compared with pasting drafts into a chat assistant, Markra keeps the AI inside the document with context and a preview step, which removes the copy-paste loop. Compared with Obsidian plugins that add AI, Markra builds the provider settings, the side panel and the command bar into the core, so there is no plugin compatibility surface to maintain. The trade is that the knowledge-management features are deliberately shallow. If your writing life is folders of .md files, occasional tables and images, and an AI pass before publishing, Markra covers that whole loop locally, and the AGPL source means you can verify the data-flow claims in docs/privacy.md yourself.
Editorial conclusion
Take Markra if you write Markdown in folders on disk and want AI involved without handing your library to a cloud editor, because the file model is plain .md, the AI actions show a preview before anything is applied, providers are yours to configure, and the Tauri desktop build stays light. Prefer Typora if you only want minimal live preview, and Obsidian if the point is a linked knowledge base rather than document writing, which is the comparison the README itself draws. Watch two things: the web editor loses spellcheck, backup and sync to browser limits, and the AGPL licence matters if you embed the editor in your own service. The release cadence, with v2.11.1 landing on 2026-09-17 and IME and table fixes in recent notes, shows a project still polishing Chinese-input and editor details that matter daily.
Frequently asked questions
Does Markra need an account or send my files anywhere?
No account is required and files stay on disk by default. Only optional WebDAV sync, remote image storage and AI requests to the providers you configure make network contact, per the README and docs/privacy.md.
What does the web editor lack compared with the desktop app?
Spellcheck, note backup and sync are unavailable on web, file tree operations are limited by browser permissions, AI requests must respect provider CORS, and export is HTML download plus browser print instead of HTML, PDF and Pandoc.
Which AI providers can Markra use?
Provider settings are multi-provider and you bring your own keys, with the 2.11.0 release adding OrcaRouter as a provider, plus support for ACP local agents on desktop.
Community notes