Tesseract Streamlit

A CLI tool that generates a Streamlit app from a running Tesseract instance. The generated app allows users to interactively submit input to the apply endpoint and optionally visualise results using custom plotting functions.

✨ Features

• 🔍 Parses Tesseract's OpenAPI schema
• ⚙️ Automatically builds input forms for the apply endpoint
• 📊 Supports user-defined plotting of inputs and outputs
• 🚀 Outputs a ready-to-run Streamlit app script
• 🧩 Modular and customizable with minimal boilerplate

🛠 Requirements

• Modern UNIX based OS (Linux or macOS), or Windows with WSL
• Python >= 3.10
• tesseract-core

📦 Installation

$ pip install tesseract-streamlit

🧰 Usage

$ tesseract-streamlit [OPTIONS] [URL] [OUTPUT]

• URL: The address to the Tesseract instance you want to interface with. Not needed with --from-image.
• OUTPUT: (Optional) The file path to write the generated Streamlit app. When omitted, the app is written to a cache file and Streamlit is launched automatically.

Quick start

If you've got a running Tesseract, you can get started fast by simply passing in its URL.

$ tesseract-streamlit http://127.0.0.1:61725

This generates the app and launches Streamlit in one step.

Or, skip serving the Tesseract yourself entirely:

$ tesseract-streamlit --from-image vectoradd

This serves the Tesseract from the given image, generates the app, and launches Streamlit, in a single command.

Full example

# Install dependencies
$ pip install tesseract-core tesseract-streamlit

# Clone Tesseract Core and build example Tesseract
$ git clone https://github.com/pasteurlabs/tesseract-core.git
$ tesseract build tesseract-core/examples/vectoradd

# Option A: Serve yourself and auto-launch the app
$ tesseract serve vectoradd --port 61725
$ tesseract-streamlit http://127.0.0.1:61725

# Option B: Let tesseract-streamlit handle everything
$ tesseract-streamlit --from-image vectoradd

# Option C: Generate a script to run later
$ tesseract serve vectoradd --port 61725
$ tesseract-streamlit http://127.0.0.1:61725 app.py
$ streamlit run app.py

⚙️ Options

🖼 Layouts

Choose how inputs and outputs are arranged with --layout:

$ tesseract-streamlit --layout simple http://localhost:48819

two-column (default)	simple
Inputs and outputs side-by-side, each scrolling independently — tweak inputs while keeping results in view.	Inputs stacked above outputs in one centered column — simpler, and easier to read with many plots.

📊 With Custom Plotting

You can optionally pass a Python file containing user-defined functions for plotting inputs and/or outputs.

$ tesseract-streamlit --user-code udf.py http://localhost:48819

The udf.py file should define functions like:

import plotly.graph_objects as go

def plot(inputs, outputs) -> go.Figure:
    """Title of the plot.

    Description of what the plot conveys. Will be displayed above the
    plot itself.
    """
    fig = go.Figure()
    # some plotly logic to visualise input_schema and output_schema
    return fig

Where we have chosen Plotly as our plotting back-end, but any supported library by Streamlit is allowed.

📁 Example

See the example README for a basic example walk-through, or simply run the following script to see the end result!

$ bash examples/vectoradd_jax/run.sh

This will open a browser window with the Streamlit UI where users can input values and visualise the response.

⚠️ Current Limitations

While tesseract-streamlit supports Tesseracts with an InputSchema formed with arbitrary nesting of Pydantic models, it does not yet support nesting Pydantic models inside native Python collection types such as:

• ❌ list[Model]
• ❌ tuple[Model, ...]
• ❌ dict[str, Model]

These types will raise an error or be ignored in the generated app.

You can however use these native collections with basic Python types, such as:

• ✅ list[str]
• ✅ dict[str, int]
• ✅ tuple[float, float]

If you would like to request support for nested models within collections in a future release, please let us know.