OpenKnowledge: A WYSIWYG Markdown Editor That Puts Claude and Codex Inside Your Notes
Beautiful, AI-native markdown IDE and LLM wiki. To get more complex, you can use the starter packs to create LLM Wikis, second brains, or more structured knowledge bases.
At a glance
- What is it?
- OpenKnowledge is a local-first, GPL-3.0 markdown IDE that turns plain folders into AI-editable wikis. It ships with MCP integration for Claude, Codex, and other harnesses, but its beta status and Node 24 requirement demand a close look before adoption.
- Who is it for?
- Adopt OpenKnowledge if you live in markdown and want a local, AI-assisted editing layer that works with your existing folder structure and your preferred agent harness. Skip it if you need a stable, non-beta tool, run Node 22 or older, or require a permissive license for commercial redistribution.
- Can I use it commercially?
- Yes, with conditions. GPL-3.0 is a copyleft licence: if you distribute software that includes it, you must release that software's source code under the same licence. Running it internally without distributing it does not trigger that obligation.
- Is it still maintained?
- Yes. The repository last received commits 1 day 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 14, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The Problem: Markdown Editing Is Split Between Plain Text and Heavy Wikis
Most engineers face a choice: edit markdown in a plain text editor with no live preview, or move notes into a proprietary wiki that locks content in a database. OpenKnowledge targets that gap. It is a markdown editor with true WYSIWYG rendering, so editing a .md file feels like editing a Google Doc or Notion page, according to the README. The tool is aimed at people who already have folders full of markdown or MDX files: existing codebases, wikis, Obsidian vaults, or plain notes. It does not require you to import data into a new format. You open a folder, and the app reads what is there. The AI angle is central: OpenKnowledge integrates with Claude, Codex, OpenCode, Pi, and other harnesses via MCP and CLI, turning the editor into a front end for agent-assisted authoring and search.
How It Works: Local Files, MCP Skills, and Git Under the Hood
The architecture is straightforward from the README. OpenKnowledge is a desktop app (macOS, Windows, Linux) and a web UI served from a local CLI. It opens any folder containing markdown or MDX. The WYSIWYG editing is the core interaction layer. For AI features, the app detects agent harnesses installed on your machine and walks you through installing an MCP server and skills. These skills are designed to give agents enriched search and document authoring capabilities. The README mentions "agentic search" for LLM wikis, second brains, and knowledge graphs. Team sharing and auto-sync are optional and powered by git/GitHub under the hood. That means your data stays in your folder, and sync is just a git push/pull cycle. The editor also includes a graph wiki link viewer, which suggests a bidirectional link feature for knowledge bases, though the README does not detail the graph syntax.
Getting It Running: Desktop Installer or Node CLI
There are two install paths. The desktop app is a standard download from openknowledge.ai/download or from the latest GitHub release. macOS Apple Silicon uses a DMG; Windows 10+ offers an x64 or Arm64 Setup installer that runs per-user with no admin prompt; Linux gets a deb for Debian/Ubuntu or an rpm for Fedora/RHEL. For servers, Intel Macs, or any platform without a desktop build, the CLI route works. It requires Node.js 24+ and git. The commands are: npm install -g @inkeep/open-knowledge, then cd into your project, run ok init to scaffold the project and wire up AI editors (Claude Code, Claude Desktop, Cursor, Codex, OpenCode, OpenClaw), and ok start --open to serve the web editor. The ok init step is notable because it does more than create a config: it actively detects and configures your installed agent harnesses.
The AI Integration Is the Differentiator, but It Depends on Detection
OpenKnowledge's core promise is that it works with whatever agent you already use. The README lists Claude, Codex, OpenCode, Pi, and "others" as supported, and says any harness can be used via MCP/CLI. That is a broad claim. The ok init command wires up specific editors: Claude Code, Claude Desktop, Cursor, Codex, OpenCode, OpenClaw. If your harness is not on that list, you are left with the generic MCP/CLI path, which may require manual configuration that the README does not document. The detection step is also a potential failure mode: if the app does not find your harness, you get no AI features out of the box. The README says the app "will walk you through installing the MCP and skills for agent harnesses detected on your computer." That implies detection is automatic, but it also means the quality of the experience depends on how well the detector knows your system.
Limitations and Failure Modes: Beta Status and Node 24
The project is clearly in beta. The latest release is v0.67.0-beta.2, with a v0.66.2 stable release just before it. That beta label matters for production use. The web app requires Node.js 24+, which is a recent LTS line; many enterprise environments still run Node 20 or 22. If you cannot upgrade Node, the CLI path is closed. The desktop app may work on older systems, but the README only lists macOS Apple Silicon, Windows 10+, and Linux with deb/rpm. Intel Macs are explicitly excluded from the desktop app, though the web app covers them. Another limitation: the README does not describe how WYSIWYG handles complex MDX or embedded HTML. It mentions "embeddable HTML and rich components" for specs, but there is no detail on what happens when you open a file with custom JSX or frontmatter. If your existing markdown uses exotic syntax, you should test it before migrating.
Alternative Approaches: Obsidian and Standard Markdown Editors
The most direct alternative is Obsidian, which also works on local markdown folders and offers a graph view. Obsidian's plugin ecosystem is mature, but its AI integration is not native; you typically install community plugins or use external scripts. OpenKnowledge bakes AI into the core via MCP, which is a different architecture. Another alternative is a plain editor like VS Code with a markdown preview extension. That gives you WYSIWYG-ish preview but no AI search or wiki graph. The difference is that OpenKnowledge treats the editor as a shell around your files and your agents, not as a standalone app. If you already use Claude Code or Codex heavily, OpenKnowledge aims to be the visual layer on top of those tools. If you prefer a non-AI editing experience, Obsidian or VS Code may be simpler and more stable.
Licensing and Upgrade Cost
OpenKnowledge is licensed under GPL-3.0-or-later. That is a strong copyleft license. If you embed or modify the code, your derivative work must also be GPL. For internal use, that is usually fine. For commercial distribution of a modified version, you must release your changes. That is a real constraint for companies that want to fork it into a proprietary product. The README does not mention any paid tier or enterprise plan, so the project appears fully free. Upgrade cost is tied to the release cadence: the project pushes beta releases frequently (two betas on the same day as a stable release). That suggests active development, but also means you may need to track releases to get bug fixes. The CLI tool is a global npm package, so upgrading is a simple npm update -g @inkeep/open-knowledge, but you should check the changelog for breaking changes, which the README does not summarize.
Editorial conclusion
Adopt OpenKnowledge if you live in markdown and want a local, AI-assisted editing layer that works with your existing folder structure and your preferred agent harness. Skip it if you need a stable, non-beta tool, run Node 22 or older, or require a permissive license for commercial redistribution. Before committing, verify that your agent harness is in the detected list (Claude Code, Codex, Cursor, OpenCode, OpenClaw) and that the MCP setup works on your OS. Also confirm that the WYSIWYG rendering preserves your existing MDX syntax, since the README does not detail edge cases.
Community notes