New Version + Feedback

[krish-adi]

I am currently working on updating the package. This is the last of changes:

1. The frontend is currently in vue. This has stopped many developers from contributing, and therefore I am migrating the frontend to react.
2. The library currently used for rendering the flows is baklavajs. I will be migrating it to react-flow, which will also improve DevX in contributing to the package.
3. Improved documentation. There is very little documentation on what is possible with the library, and expect this to be improved.
4. Multimodal rendering on the UI. Currently the blocks can only render text based outputs using the display option. This will be extended to include documents, image and audio.

If you have more suggestions on how to improve this library, please add them on here.

[zabrewer]

@krish-adi - first I hope you are well.

Are the new changes going to break the API or way that barfi is called? Do you expect any breaking changes?

Will streamlit still be used?

I have a project I'm close to releasing that uses barfi as a dependency.

[krish-adi]

Hey @zabrewer !

The API will not change, and will still use Streamlit.

What will mostly change in the upcoming release is the UI. But the blocks and their interaction in the UI will remain the same.

[krish-adi]

I will mark a few issues as enhancements, that I'll look into deploying into the next release.

[equidna]

Could you add support for jupyterlab? It would be incredible to manage the functions workflow inside a jupyter notebook (maybe something similar to makinarocks link but FOSS).

[krish-adi]

hey @equidna , a jupyterlab extension is something I have been thinking about, so will put it in the roadmap. I haven't heard about Makinarocks, could you share an image and a description of what you are referring to and/or expecting to get?

[equidna]

@krish-adi The makinarocks link allows to use any notebook cell as a node (link with images: https://link.makinarocks.ai/). Unfortunately the code is closed source and the user needs to create an account to use it.

There is another similar solution called elyra pipelines (link with images: https://elyra.readthedocs.io/en/latest/getting_started/overview.html#ai-pipelines-visual-editor). This tool allows to use any notebook as a node, but is hard to schedule pipeline runs without kubernets or apache airflow and I think it doesn't have an updated version for jupyterlab 4.

My idea would be to run barfi as the output of a cell and instantiate functions of a notebook or cell as nodes (allowing for easier prototyping and testing different workflows). I believe this would allow to parameterize code workflows and code reuse. Instead of just parameterize input values, like the usual papermill solution (https://github.com/nteract/papermill).

[ANGJustinl]

Could you add groups setting, like categorize nodes into diffrent groups?