Zett
Install

Catch the thought
before it is gone.

Zett is a private workspace where one line you jot down becomes a card, an article, a slide deck, or a paper — shaped with an assistant that keeps everything on your machine.

Install Zett Read the docs

Python 3.10–3.14 · macOS, Linux, Windows · no account, no telemetry

One line in. Something kept, out.

AI workspace DeepSeek Medium

What should we remember? Type a rough thought, or drop in a file.

create_artifact draft

Card · idea unsaved draft

Local-first beats sync for personal notes

A file on disk survives the company that made the app.

local-firstownership
All Cards Articles Slides PDFs
Card · idea Local-first beats sync for personal notes local-first · ownership
Article How Zett keeps a long conversation cheap compaction
Slides Ten minutes on local-first notes talk
Paper CRDT-based note sync sources kept

  1. Write one line.

    A passing idea, a rough note, a question you cannot answer yet. Type it the moment it arrives; the workspace is already open.

  2. Ask for the shape.

    “Turn this into a card.” “Draft a deck.” The assistant reads what you attached and writes a first version, in the open, so you can argue with it.

  3. Keep it, on purpose.

    Nothing is published until you say so. Review the diff, tag it, save it — and it joins a library that stays on your disk.

Everything you keep, in one library.

Ask for a card, an article, a deck, or a paper. Nothing is published until you save it, and everything you save is searchable, tagged, and still editable.

Card · idea draft

Local-first beats sync for personal notes

A file on disk survives the company that made the app. Sync is a feature you can add later; ownership is not.

local-firstownership
Article saved · v2

How Zett keeps a long conversation cheap

Compaction replaces old turns with a checkpoint while the raw log stays on disk, so a long thread never costs a fortune to continue.

compactioncontext
Slides saved · 8 slides

Ten minutes on local-first notes

A deck built from the same conversation, presented full screen and exported to PDF when the room asks for one.

talkteam
Paper saved · 12 pages

CRDT-based note sync

Ask for a paper and it compiles one, keeping the sources beside it so you can edit, rebuild, and export it whenever you like.

sources keptpdf

Your notes never leave your machine.

Reachable, and patient.

Send it a photo from your phone, or hand it a schedule and read the transcript later. Both write to the same library, on the same machine.

Ask from your phone

Connect a chat platform and every conversation keeps its own thread. A photo, a voice note, or a file arrives as a file, and the assistant writes to the library while you are away from the desk.

You
Zett
AI workspace

Or let it run without you

Put a prompt on a schedule and read the transcript afterwards. Each run gets a fresh session, so an unattended task never lands in your conversation list.

Weekly reading digest Mon 09:00 · Asia/Shanghai

Summarize what I saved this week into one card.

Enabled Run now
  • New session started00:00:02
  • list_assetscreate_artifact00:00:19
  • Card saved · 6 sources00:00:42
Weekly reading digest Card · saved to the library · next run in 7 days

Let your coding agent file the result.

Zett's artifacts and assets are plain HTTP, and exposed over MCP. Point Claude Code or Codex at the API and what it just finished lands in your library — a card, an article, or an uploaded file.

Claude Code idle Codex idle
Zett
AI workspace

    The same endpoints are exposed over MCP, so an agent that already speaks it needs no new code.

    Install it like a tool, not a service.

    One command, and it runs on your machine with everything it knows in one folder. Nothing to sign in to, nothing to keep running in the cloud. 

    $ uv tool install zett$ zett start        # 127.0.0.1:6280, in the background
    Read the install steps Browse the docs

    Then connect a model.

    Zett brings no model of its own. Add a provider once — the key is encrypted and stays on your machine — and every conversation has something to think with.

    Settings No model yet

    No provider configured yet. Zett cannot answer anything until you add one.

    Temperature 0.2
    API mode Chat Completions Responses API Responses mode enables provider-hosted tools such as web search.
    Credentials are encrypted in your local database. Enabled Add provider

    Configured providers

    DeepSeek · deepseek-chat key stored on this machine · enabled
    + New provider
    1. Start it once.zett start serves the workspace in the background; zett status reports it.
    2. Add a provider.Name it, pick a model, paste a key.
    3. Then just talk.The provider is ready for every conversation.