What nooklet is

nooklet is an outliner for one person with several devices. You write in nested bullets, link pages with [[wiki links]] and #tags, and walk the graph backwards through references. If you have used Logseq, Roam or Obsidian's outliner, you know the shape.

It is open source (MIT), self-hosted, and early. The author uses it daily. Expect rough edges and breaking changes before 1.0.

Who it is for#

• You keep a daily journal and a pile of linked pages, and you want them on your own disk.
• You use more than one device and want them to agree without a cloud account.
• You want an AI agent (Claude Code, Cursor, anything that speaks MCP) to search, read and edit your notes block by block, instead of pasting whole pages into a chat.
• You have a Logseq file graph and want to bring it along.

What makes it different#

Local-first. Every device holds a full SQLite copy of the graph and works offline. Edits queue on the device and sync when it reconnects. A server is optional: one machine with no server is a normal way to run nooklet.

Agents are first-class users. One registry of operations drives the HTTP API, the OpenAPI document, the MCP tools and the typed client. An agent gets the same operations the app uses: search (keyword, semantic or hybrid), read a page with stable block ids, append markdown, edit an exact substring inside one bullet, run an atomic batch, and undo that batch. With a separate permission it can also see and drive a window you have open. See For AI agents.

A markdown mirror you can walk away with. The server writes every page to a plain markdown file with a stable ^id on each block. Logseq and Obsidian can open those files, and grep and git work on them. The files are a copy: SQLite holds the truth and the op log carries sync. See How it works.

Everything optional is off. Sync needs a device you choose to pair. AI access needs a token you choose to mint. Semantic search needs an embedding model you choose to run. Skip all three and you have a local outliner with a markdown mirror.

Non-goals#

The scope stays small on purpose. These are not planned:

• flashcards or spaced repetition
• kanban boards, whiteboards, or other "views" of the graph as a canvas
• real-time co-editing with cursors
• multi-user accounts and permissions (one owner, many devices)
• end-to-end encryption (it would block server-side search, embeddings and MCP)
• Logseq's Datalog {{query}} blocks, org-mode, LOGBOOK time tracking
• a hosted service or a plugin marketplace

If you want one of these, open an issue and argue for it. A clear "no, because" beats an issue that sits open forever.

Platforms#

Next: Features for a tour, or Getting started to run it.