OpenKnowledge is a local Markdown editor that aims to turn a folder of notes into a wiki readable by humans and usable by AI agents. After a real test on a small security/DevOps knowledge base, my verdict is straightforward: worth trying if you already live in Markdown, but not quite the smoothest tool for everyone just yet.
What is OpenKnowledge?
OpenKnowledge positions itself as a blend of Notion, VS Code, and a wiki built for AI agents. The idea isn’t to replace Git or your Markdown files, but to put a clean interface on top of an existing folder.
In practice, the product can:
- open a folder containing Markdown or MDX files,
- display a local web interface with a file explorer,
- create starter packs, for example a three-layer knowledge base,
- prepare MCP integrations and skills for Claude, Codex, Cursor, opencode, and other agents,
- keep content as local files, making it fully versionable with Git.
This matters: OpenKnowledge is not a SaaS notes app. In my test, everything stayed in a local folder. That’s genuinely reassuring for freelance work, client projects, audits, or infrastructure documentation.
| Element | Tested detail |
|---|---|
| Product | OpenKnowledge |
| npm version | 0.29.1 |
| GitHub repo | inkeep/open-knowledge |
| Inspected commit | bd0757e4 |
| Required runtime | Node.js 24 or higher |
| Interface | Local web, plus macOS app announced |
| Data | Local Markdown files |
| License | GPL-3.0-or-later |
Installing OpenKnowledge
I started by cloning the GitHub repo into /tmp, reading the README, the package.json manifests, and the package structure. The project is a Bun monorepo with packages for the CLI, the React app, the server, the core, and a native component.
The basic documented commands are:
npm install -g @inkeep/open-knowledge
cd votre-projet
ok init
ok start --open
In my test environment, Node.js 20 was installed by default, while the project requires Node.js 24. I installed Node.js 24 in a temporary directory, then installed the OpenKnowledge 0.29.1 npm package to run the product.
I also tried building from the source clone. The core, server, and web app compiled fine with Bun once Node 24 was available. The full CLI build from the repo stalled on packages/native-config, because the native component requires Rust and cargo wasn’t present in the container. That’s not a functional bug in the application, but it’s real friction if you want to contribute or rebuild the entire package from scratch.
To cover the AI use case, I exported OpenAI/OpenRouter proxy variables before launching:
export OPENAI_BASE_URL=http://172.17.0.1:8790/v1
export OPENAI_API_KEY=sk-proxy
export OPENAI_API_BASE=http://172.17.0.1:8790/v1
export OPENROUTER_BASE_URL=http://172.17.0.1:8790/v1
export OPENROUTER_API_KEY=sk-proxy
In this specific test, OpenKnowledge didn’t need to call an LLM directly to display the wiki. Its AI angle is mainly in the MCP integrations, the skills, and making the folder understandable by agents.
OpenKnowledge in Practice
To avoid testing on an empty setup, I created a small freelance security/DevOps knowledge base:
- a SOC wiki README,
- an incident note about noisy Prometheus alerts,
- a Prometheus triage runbook with a wiki link to the incident.
I then initialized OpenKnowledge without writing a global MCP config:
ok --cwd /tmp/ok-case init --no-mcp --local-only --json
The JSON output confirmed that the project had been detected, that three content files already existed, and that MCP registration had been skipped:
{
"projectRoot": "/tmp/ok-case",
"contentDir": ".",
"contentFileCount": 3,
"mcpAction": "skipped-flag"
}
Then I added the knowledge-base starter pack:
ok --cwd /tmp/ok-case seed --pack knowledge-base --root knowledge -y
What that produced: 17 starter pack entries, a knowledge/log.md file, knowledge/articles, knowledge/research, and knowledge/external-sources folders, frontmatter files, and a local OpenKnowledge skill. The preview command then showed 5 Markdown files tracked by the watcher, including my notes and the starter pack.
ok --cwd /tmp/ok-case preview
Useful output:
Content:
Found 5 markdown files in ./
Sample: .ok/skills/open-knowledge-pack-knowledge-base/SKILL.md, README.md,
incidents/2026-07-11-alerting.md, knowledge/log.md,
runbooks/prometheus-triage.md
I then started the local server:
ok --cwd /tmp/ok-case start --host 127.0.0.1 --port 5337 --ui-port 5338 --mode browser
The web interface started successfully, with an editor at http://127.0.0.1:5338 and an API at http://127.0.0.1:5337. In the headless browser, OpenKnowledge displayed the ok-case workspace, the starter pack templates, the incidents folder, my Prometheus incident, and global search.
The most convincing part of the test: search found my Markdown files and displayed readable excerpts. The runbook and the incident stayed as plain files, but the interface makes them navigable without imposing an external database.
What I Like About OpenKnowledge
First, the product respects the Git mindset. Content stays in Markdown with an understandable structure. For infrastructure documentation, client procedures, runbooks, or research notes, that’s a genuine advantage.
Second, the knowledge-base starter pack provides a solid framework. It pushes you to separate raw sources, research, and finalized articles. That’s exactly the kind of discipline that most internal wikis lack.
I also appreciate the agent-friendly positioning. The .ok files, skills, and MCP give a clear direction: a human writes and reviews in a clean interface, while an agent can search, create, consolidate, or manipulate notes using explicit conventions.
Finally, the interface is pleasant. The first screen isn’t overwhelming, search is fast on my small dataset, and templates are visible right from the home screen.
The Limits of OpenKnowledge
The first limit is installation for non-developer profiles. The README calls for Node.js 24, Git, and a CLI. For general users, that’s already a significant barrier. On macOS, the desktop app could reduce that friction, but I didn’t test it in this Linux container.
The second limit is the full source build. The monorepo is well structured, but rebuilding the entire package requires Bun, Node 24, and, for the native component, Rust. In a CI environment or a minimal container, you need to prepare the ground carefully.
Third limit: the AI isn’t magic on first launch. OpenKnowledge sets the stage very well for agents, but if you’re looking for a button that automatically generates all your documentation, that’s not what I observed here. The product is more of a solid documentation orchestration layer.
Last point: initializing without MCP correctly skipped MCP registration, but the seed still installed skills in the project for several agents. That wasn’t a problem in my test, but I’d appreciate even more explicit options for locked-down environments.
Does OpenKnowledge Actually Work?
Yes, for the core of its promise: open a local Markdown folder, initialize it, add a knowledge base structure, start a web UI, and browse the notes.
What I actually verified:
- installing the npm package 0.29.1 under Node 24,
- inspecting and partially building the source repo,
- initializing a local Markdown project,
- generating the
knowledge-basestarter pack, - starting the local web server,
- displaying the workspace in headless Chromium,
- searching for a domain-specific term (
Prometheus) with relevant results.
What I didn’t validate end to end: the macOS app, team sharing via GitHub, multi-user collaborative editing, and a complete loop with an external agent calling MCP tools to rewrite the wiki.
Should You Adopt OpenKnowledge?
My verdict: try it, especially if you already have a Markdown and Git culture.
For a freelance security/DevOps professional, OpenKnowledge can help structure runbooks, audit notes, architecture decisions, incident logs, and technical research. The local format is a real advantage when content is sensitive.
I wouldn’t recommend it yet as an immediate Notion replacement for a non-technical team. For a team of engineers or an independent developer who wants a local wiki that’s readable by agents and versionable, though, it looks promising.
Final verdict: adopt it in a lab or personal project, try it on a small client wiki, and wait a bit longer before migrating an entire critical documentation base.
FAQ
Does OpenKnowledge replace Obsidian?
Not exactly. OpenKnowledge is more focused on technical wikis and AI agent workflows, while Obsidian remains very strong for personal note-taking and plugins.
Does OpenKnowledge require an OpenAI key?
Not to open and browse a local Markdown wiki. Agent-oriented features may depend on an external tool and a model, but the core tested here works without any direct LLM calls.
Is OpenKnowledge ready for a DevOps team?
Yes for a controlled trial on runbooks or a project knowledge base. For a broader rollout, I’d first check Git rules, access permissions, backups, and the MCP workflows used by agents.