The Problem

Technical teams that need a WYSIWYG markdown experience for knowledge‑base authoring currently have to choose between generic editors (Notion, Google Docs) that offer no AI‑augmented authoring, or developer‑focused tools (VS Code, Obsidian) that lack rich‑text styling and out‑of‑the‑box LLM integration. The result is duplicated effort: authors maintain separate docs and code, and AI harnesses must be wired manually for each repository.

What This Does

OpenKnowledge is a markdown IDE that doubles as an LLM‑wiki with integrated AI editing, git‑backed sharing, and a desktop/web cross‑platform client.

  • Core UI lives in packages/app/ – the React/Next.js app (packages/app/src/App.tsx is the root component, rendered via packages/app/index.html).
  • Editor extensions are under packages/app/src/editor/: asset-dispatch/index.ts, clipboard/index.ts, image-upload/index.ts handle file I/O and AI‑mediated asset flow.
  • AI harnesses (Claude, Codex, OpenCode, etc.) are wired through the MCP/CLI infrastructure exposed by packages/cli/. The CLI scripts in .github/scripts/ (e.g., cla-gate.mjs, write-back.mjs) manage skill registration and artifact persistence.
  • Document store is markdown/MDX files under docs/ (249 files) and the packages/ monorepo; changes are synced via git/GitHub using the CI pipelines in .github/workflows/.
  • Styling uses Tailwind (docs/tailwind.config.ts) and the Biome plugin set in biome-plugins/ for linting/format checks.
  • Configuration is provided by .env.example at root and docs/.env.example; required keys (e.g., API tokens for LLMs) are injected at runtime.

The repository contains 5 596 files (TypeScript 4 058, TSX 1 025, Markdown 86, JSON 84, YAML 32, Shell 20, HTML 13, Rust 7). Test coverage is substantial: 2 879 test files discovered, and 267 documentation files are present.

How It Is Wired

  1. Entry point – packages/app/index.html loads the React app; the first render is packages/app/src/App.tsx.
  2. App → editor modules – App.tsx imports asset-dispatch, clipboard, and image-upload to give the editor drag‑and‑drop, copy/paste, and image insertion capabilities.
  3. AI editing – When a user invokes an AI skill, the request is routed through packages/cli/ scripts (e.g., cla-gate.mjs validates the skill manifest) and then to the selected LLM via MCP. The returned markdown is written back through write-back.mjs, which updates the underlying .md/.mdx file and commits the change via the GitHub Actions workflows (.github/workflows/release.yml, .github/workflows/write-back.yml).
  4. Sync & sharing – Git‑based sync is handled by the CI pipelines; the share-contract-* scripts (share-contract-reader-gate/action.yml, etc.) expose a contract for team‑wide document sharing.
  5. Build & lint – biome.jsonc and biome-plugins/ enforce code style; pnpm (package.json at root) resolves dependencies; Next.js config (docs/next.config.ts) builds the web UI.

No structural red flags were detected: tests, CI, license, and lockfile are all present where expected.

How To Use It

StepCommand (from README)What it does
Install desktopDownload from <https://openknowledge.ai/download> or latest release (github.com/inkeep/open-knowledge/releases/latest)Provides macOS DMG, Windows exe, Linux deb/rpm.
Install web CLInpm install -g @inkeep/open-knowledgeInstalls the ok CLI binary.
Scaffold a projectcd your-project && ok initGenerates ok.config.json, wires default MCP/skills, creates a starter .md vault.
Start web UIok start --openSpins a local dev server and opens the editor in the default browser.
Add AI skillsEdit ok.config.json (or use ok config add …) – specifies which LLM harnesses (Claude Code, OpenCode, etc.) are available.
Enable git syncEnsure a GitHub remote is added; the write-back.yml workflow will auto‑commit markdown changes.

Environment variables (API keys, etc.) are expected in .env.example at the repo root and docs/.env.example; copy these files and fill in your keys before running ok start.

Real‑World Use

A product team maintaining an internal knowledge wiki can clone the repo, run ok init, and immediately have a searchable, WYSIWYG markdown canvas. Authors write meeting notes or specs in the editor; the team enables the Claude skill, so an LLM can instantly rewrite a section, summarize a paragraph, or fill a template. Because changes are committed to git, the wiki stays versioned and diff‑able. Team members share the vault via GitHub, and the share-contract gates enforce access control without additional infrastructure.

Code Health & Issues

  • Measured analysis (static, not opinion):
  • 2 879 test files detected across the monorepo.
  • 267 documentation files present (.mdx, CHANGELOG.md, etc.).
  • No missing license or lockfile; LICENSE and package-lock.json/pnpm-lock.yaml are present.
  • CI pipelines (GitHub Actions) cover build, test, release, and share‑contract gates.
  • Potential SDLC observations:
  • The packages/ directory is large (5 151 files) – cloning or CI runs may be slow on low‑end machines.
  • Some workflow scripts (e.g., bridge-public-pr-to-monorepo.mjs) are complex; a newcomer may need to review the .github/scripts/ directory to understand PR‑bridging logic.
  • No explicit Dockerfile in the root; containerised deployment would require a custom Dockerfile or using the provided installers.

The Bottom Line

OpenKnowledge delivers a polished, AI‑native markdown editor that bridges the gap between a rich text workspace and an LLM‑augmented knowledge base. Its strength lies in the integrated MCP/CLI skill system, git‑backed sharing, and cross‑platform availability. The main trade‑off is the size of the monorepo and the need to configure LLM API keys before full AI editing can be used. It is well‑suited for teams that want a unified wiki/notes platform with built‑in agent assistance and are comfortable with a moderately large codebase.