jupyterlab-starchat-extension

Note You will need a StarChat endpoint. See starchat-service

A JupyterLab extension that implements the StarChat coding assistant. Uses Fable tooling.

The following query string parameters enable functionality:

• endpoint=xxx specifies a url for the StarChat endpoint (e.g. https://yourdomain.com/starchat) the extension will not work properly without this
• log=xxx specifies a url for a logging endpoint (e.g. https://yourdomain.com/log)
• id=xxx adds an identifier for logging

The log data format is {username: string, json: string}.

NOTE: This plugin requires jupyterlab <= 1.2.6, so if you have a higher version (e.g. 2.X) you will need to execute conda install jupyterlab=1.2.6 or similar for pip. The conda environment specification provided in environment.yml will work as-is (e.g. mamba env create -f environment.yml) or can be used as a template.

Installation

jupyter labextension install @aolney/jupyterlab-starchat-extension

Updating to latest version

jupyter labextension update @aolney/jupyterlab-starchat-extension

Development

This is based on my personal preferences. For more options, see the extension development guide.

Prerequisites

• JupyterLab
• Fable
• An F# editor like Visual Studio Code with Ionide
• Chrome

Initial install and after library adds

npm install
mono .paket/paket.exe install
npm run build

Terminal A in VSCode

jupyter labextension install . --no-build
npm run watch

This will watch your F# code and trigger builds of index.js.

If you prefer not to trigger builds using a watch, you can npm run build every time you want a new build.

Terminal B in VSCode

jupyter lab --watch

This will watch your extension and trigger builds of it.

Even with this watch, you still need to refresh your browser during development.

Project structure

npm/yarn

JS dependencies are declared in package.json, while package-lock.json is a lock file automatically generated.

paket

Paket

Paket is a dependency manager for .NET and mono projects, which is designed to work well with NuGet packages and also enables referencing files directly from Git repositories or any HTTP resource. It enables precise and predictable control over what packages the projects within your application reference.

.NET dependencies are declared in paket.dependencies. The src/paket.references lists the libraries actually used in the project. Since you can have several F# projects, we could have different custom .paket files for each project.

Last but not least, in the .fsproj file you can find a new node: <Import Project="..\.paket\Paket.Restore.targets" /> which just tells the compiler to look for the referenced libraries information from the .paket/Paket.Restore.targets file.

Fable-splitter

Fable-splitter is a standalone tool which outputs separated files instead of a single bundle. Here all the js files are put into the lib. And the main entry point is our index.js file.

Imports

Because Jupyter uses Typescript, we can use ts2fable to generate strongly typed imports of Jupyter's JS packages. Unfortunately these are a bit huge and the conversion is messy. Jupyter.fs contains the minimal imports for this demo repo, and these imports were tweaked out of the files in the ts2fable-attempt directory.