OpenKnowledge: Beautiful AI-native markdown editor

My notes

Summary

OpenKnowledge is a local-first, open-source markdown editor and knowledge base designed as shared persistent memory for AI coding agents. It positions itself as the “agent brain”, a single set of plain markdown files that both humans and agents (Claude, Cursor, Codex) read from and write to, synced through git so history and ownership stay with the user.

Key Insight

• Core thesis: agents lose context inside chat threads. The fix is a persistent shared file store both humans and agents edit in the same loop, not a thread, but a wiki of plain markdown.
• Plumbing under a Notion-like UI: rich blocks (callouts, collapsibles, tabs, Mermaid diagrams, images, embedded video, tables, code, drop-in apps) but the underlying storage is plain markdown you own.
• Built-in agent tooling: agents get navigation/edit/grow operations out of the box, plus embeddings + hierarchical RAG so they find the right content rather than dumping the whole base into context.
• Integration surface: plugs into Claude, Cursor, Codex and other agents, meant to be agent-agnostic memory, not tied to one IDE.
• Ownership/privacy stance: local-first, runs on your machine, self-hostable, git-synced, community-built. The differentiator vs. cloud knowledge tools is that the files stay yours and auditable.
• Use cases pitched: shared agent memory, team wiki (skills/docs/notes auto-synced), and specs/PRDs/roadmaps living next to code edited by humans + coding agents together.
• Caveat: this is landing-page copy, no benchmarks, pricing, or maturity signals. Treat as a pattern to evaluate, not a validated tool.