jupyterlab-dat

An experimental approach for publishing/subscribing to streams of notebook contents in JupyterLab over the dat peer-to-peer network This is not real-time collaboration.

Usage

In a binder or locally

Publish

• Open a notebook
• Click the icon in the notebook toolbar
• Click PUBLISH
• Copy the dat:// url, send it to someone else with jupyterlab-dat installed
• Change the notebook

Subscribe

• Open a new Untitled notebook
• Click the icon in the notebook toolbar
• Paste a dat:// url into the text box above SUBSCRIBE
• Click SUBSCRIBE
• Watch the notebook change

Chat

• Click the ⬡ icon in the Lab sidebar
• Click my dat
• Enter a name for yourself
  • Optionally, click publish
• Open dat chat
• If you've published or subscribed a dat, you can chat to other publishers/subscribers
  • Or paste a dat:// url into the text box
• Click CHAT

Features

• two-or-three click publish/subscribe
• optional metadata sharing
• live view of peer count and version
• active cell following (enabled by default)
• markdown rendering (enabled by default)
• trust cells (enabled by default)
• single-link sharing, e.g.

  https://mybinder.org/v2/gh/{:repo}/{:branch}?urlpath=lab/tree/dat/notebook/{:dat-hash}
  

  • opens an Untitled notebook
  • starts subscribing
  • closes sidebars
  • activates single-document and presentation mode

How it works

• a publisher
  • joins a swarm
  • creates a archive
  • stores changes to a notebook as a stream of small files a la nbexplode in a dat
  • send subscribers a 64-character key to subscribers over an out-of-band channel e.g. chat
• a subscriber
  • joins a swarm
  • opens a notebook, and subscribes with the provided key
  • replicates the archive
  • pushes the small files as received to their notebook
• publishers and subscribers run a discovery-swarm-web process to help find peers in the swarm
  • provided via jupyter-server-proxy

Limitations

• archives are created in-memory, with no persistence, and reference transient JupyterLab cell ids
  • if a publisher closes their page, that's it (though existing subscribers will keep their notebooks)
• widgets, bokeh, or anything else comm-based, don't work at all

Try locally

Needs a conda installation, e.g. Miniconda or Anaconda

(base) conda env update
(base) source activate jupyterlab-dat
(jupyterlab-dat) bash postBuild
(jupyterlab-dat) jupyterlab