LlamaViz

Turn your LlamaIndex Workflows into beautiful UIs!

Installation

Just do

pip install "git+https://github.com/run-llama/llama_viz.git"

and you are in business.

Quick start

To build a UI over an existing workflow is a two step process:

1. Wrap your workflow in a Viz instance.
2. Call the Viz.run() method to start the UI.

Tip

Using custom Start and Stop events in your workflow will let you take full control over the generated UI.

Example code:

# Put this code in a script, for example 'image_workflow.py'
from llama_index.core.workflow import Context, StartEvent, StopEvent, Workflow, step
from openai import OpenAI
from pydantic import HttpUrl

from llama_viz import Viz


class QueryEvent(StartEvent):
    """The `query` parameter will be shown as an input text box in the UI."""

    query: str


class ImageDrawn(StopEvent):
    """The `image` parameter will be shown as an image coming from an url in the UI."""

    image: HttpUrl


class ImageDrawWorkflow(Workflow):
    @step
    async def chat(self, ctx: Context, ev: QueryEvent) -> ImageDrawn | None:
        client = OpenAI()
        response = client.images.generate(
            model="dall-e-3",
            prompt=ev.query,
            size="1024x1024",
            quality="standard",
            n=1,
        )

        # Get the URL of the generated image
        image_url = response.data[0].url or ""
        return ImageDrawn(image=HttpUrl(image_url))


if __name__ == "__main__":
    # Just wrap the workflow in a Viz object, no additional code needed.
    ui = Viz(ImageDrawWorkflow())
    # Run the UI, the workflow will run in the background.
    ui.run(debug=True)

To build and run the UI, from a terminal:

$ python image_workflow

If there are no errors, point your browser to http://127.0.0.1:8050/.