OpenKnowledge

WYSIWYG markdown knowledge-base editor that ships a 21-tool MCP server and skills for coding agents.

AdaptableOperable Business appDesktop app Hybrid TypeScript MCP server License: GPL-3.0

OpenKnowledge from Inkeep is a desktop and local-web markdown editor for LLM wikis, specs, and notes over any folder of .md files. Its installer wires an MCP server and skills into Claude Code, Codex, Cursor, and other harnesses; the server exposes search, link-graph, lint, history, and polymorphic write/edit/move tools with git-backed versioning.

Use it when

you keep specs and a team wiki as markdown in git and want agents to search, lint, and edit them with history
you want an Obsidian-like editor that installs its own MCP and skills into Claude Code and Codex

Gotchas

GPL-3.0
Semantic search sends content to the embeddings provider; off by default
Every MCP call takes a cwd; a globally registered server needs it unless the client advertises one root
Web app needs Node.js 24+ and git

Last activity Sep 3, 2026. Fit: adaptable 2, operable 3.

Editions

Free · Free open source

Clonegit clone https://github.com/inkeep/open-knowledge
DownloadDownload
Installnpm install -g @inkeep/open-knowledge && ok init

For agents: JSON · Markdown

Adaptable: what is yours to change

Make it yours

Config only15 things to edit

Three YAML layers: ./.ok/config.yml (committed), ~/.ok/global.yml (user), ./.ok/local/config.yml (machine). Keys cover content.dir, contentRules, validation, appearance, autoSync, server, editor, agents.autoApproveOkTools, search.semantic. Env overrides are OK_* (OK_BIND, OK_EXTERNAL_URL, OK_ALLOW_EXTERNAL, OK_MCP_AUTOSTART, OK_EMBEDDINGS_API_KEY). Skills are SKILL.md bundles under .agents/skills/. The repo .env.example only sets VITE_PORT and OTEL_* for development.

Env vars: OK_BIND OK_EXTERNAL_URL OK_ALLOW_EXTERNAL OK_LANG OK_LOG_LEVEL OK_OPEN_BROWSER OK_MCP_AUTOSTART OK_EMBEDDINGS_API_KEY VITE_PORT OTEL_SDK_DISABLED
Config files: .ok/config.yml ~/.ok/global.yml .ok/local/config.yml ~/.ok/secrets.yml
Prompt files: .agents/skills/*/SKILL.md
Accounts you need: GitHub (optional, for team sync)OpenAI or a self-hosted embeddings endpoint (optional, semantic search)

Source

github.com/inkeep/open-knowledge
GPL-3.0 license
Last commit Sep 3, 2026
3,964 stars

Operable: what an agent can drive

Interface

OfficialMCP serverpublished by the builderlocal

How an agent can drive an Operable Listing, in five levels: native (built for the agent as the primary interface), official (the Builder ships an agent interface), wrapped (a third party wraps the software's API), config (an agent can only drive it by editing files), gui (screen control only). An interface is official only when it is published by the same organization that ships the software. GUI-level Listings are allowed and sorted last.

Docs: openknowledge.ai/docs/reference/mcp
Tools: exec search links lint audit history skills config palette preview_url share_link write edit delete move install import checkpoint restore_version conflicts resolve_conflict

Reach

Partial coverage. 8 of 10 jobs reachable through the stated interface.

Job	Through the interface	Evidence
Read and search the knowledge base	Reachable	exec (fs-direct reads with backlinks) and search (BM25 plus optional semantic)
Create, edit, move, and delete documents, folders, and assets	Reachable	write, edit, delete, move tools polymorphic over document, folder, template, skill, asset
Wiki-link graph and backlinks	Reachable	links tool with backlinks, forward, dead, orphans, hubs, suggest
Content rules and lint	Reachable	lint and audit tools; lint fix: true auto-fixes
Version history, checkpoints, and restore	Reachable	history, checkpoint, restore_version
Manage and import agent skills	Reachable	skills, install, import tools
Git and GitHub sync, conflict resolution	Reachable	conflicts and resolve_conflict; auto-sync itself is configured in autoSync.* config
Team share links	Reachable	share_link tool
Resolve comment annotations	Undetermined	exec returns unresolved comments; no explicit resolve tool seen in the reference
In-app AI chat and slides	Not reachable	human UI features; no MCP tool

Code stack

Languages: TypeScript
Frameworks: ElectronViteHocuspocus
Package manager: npm
Database: none
Hosting target: local
Built: Unknown (recorded, never ranked)

Model stack

Providers: OpenAI-compatible embeddings (OpenAI, Ollama, vLLM, LM Studio)
Models: text-embedding-3-small
Keys you need: OK_EMBEDDINGS_API_KEY
Prompts live in: .agents/skills/

Skills

Server ships skill packs (knowledge base, codebase wiki, entity vault, plain notes) and a Claude Code plugin that agents install to use the knowledge store.

Kind: bundled
Paths: packages/server/assets/skills/*/SKILL.md packages/server/assets/skills/packs/*/SKILL.md packages/plugin/.claude-plugin/plugin.json

Evidence and confidence

Field	Where it was seen	Confidence
formFactor	README: desktop app for macOS, Windows, Linux; also a local web UI from the CLI	85%
executionModel	agent integrations and opt-in embeddings; editor works without a model	75%
operability	docs/content/reference/mcp.mdx in the repo, 21 tools, published by Inkeep	95%
effort	openknowledge.ai/docs/reference/configuration: YAML config layers and OK_* env vars	85%
editions	README: 'Private, local, and free'; no paid edition found in docs	75%
active	pushedAt 2026-09-03; Show HN 2026-06-25 with 381 points	–
provenance	AGENTS.md present; no statement on how it was written	30%
reach		90%

Open questions

Does Inkeep sell a hosted or team edition of OpenKnowledge, or only its separate Inkeep products?
Which providers and keys does the in-app AI chat use?
Is 'official' or 'native' the better level given agents are a first-class co-editor?

Sources

github.com/inkeep/open-knowledge sourcing:notes-knowledge:community
openknowledge.ai/docs/reference/mcp analysis
openknowledge.ai/docs/reference/configuration analysis