Open Knowledge

Open Knowledge is an agent-native knowledge platform where humans and AI co-create, co-edit, and co-maintain knowledge.

Core principles

• Markdown files in git as the canonical substrate -- not a database, not a proprietary format.
• Agent-agnostic -- no LLM inference in the OSS core. All intelligence comes from external agents (Claude Code, Cowork, Cursor, Codex) via MCP tools.
• Everything branchable -- all meaningful state is files in git. Drafts, experiments, and proposals are branches. No separate state management.
• Rich editing -- Obsidian-grade WYSIWYG with source toggle, not a terminal-grade text editor.
• Local-first -- single-player IC to start, with CRDT architecture that gives a clear path to real-time collaboration.

Tech stack

Positioning

"Obsidian, but agent-native and collaborative."

The P0 audience is individual contributors -- developers and knowledge workers using Claude Code or Cowork -- who need a knowledge base their agent can reason over AND that they can edit with a rich experience.

Quick start

npx @inkeep/open-knowledge clone https://github.com/company/docs  # Clone + start
npx @inkeep/open-knowledge init     # Or: scaffold on an existing project
npx @inkeep/open-knowledge start    # Hocuspocus collab; auto-spawns the editor at http://localhost:3000
npx @inkeep/open-knowledge preview  # Show which files the watcher will track

clone is the fastest path from a GitHub URL to editing -- it clones the repo, scaffolds the project, and starts the editor. Changes sync with GitHub automatically. For existing local projects, use init + start instead.

After init, open your project in an MCP-compatible editor (Claude Code, Cursor, VS Code, or Windsurf) and approve the open-knowledge server. The agent can start working immediately — start is optional. When the agent calls its first MCP tool and no collab server is running, ok mcp detach-spawns ok start as a sibling process so resumption is zero-ceremony.

Documentation