Managing Project Parameters¶

Bauplan parameters let you define and manage variables that we can pass programmatically to your pipelines — things like prompts, time ranges, flags, constants, and API tokens. They’re stored in bauplan_project.yml and versioned with your code, so your logic stays declarative, reproducible, and easy to test. Parameters in your bauplan_project.yml become defaults but can always be overridden explicitly at runtime.

Parameters can be:

• Strings (str)

• Numbers (int, float)

• Booleans (bool)

• Secrets (encrypted API keys or credentials)

You can think of them as environment variables passed as arguments into your Models, but safer, versioned, and scoped to your Bauplan project.

Why Use Parameters?¶

Parameters help you:

• Customize pipeline behavior without modifying code

• Pass secrets securely at runtime

• Reuse the same logic across different configurations, inputs, or prompts

They’re particularly useful when calling APIs, toggling features, or injecting user-defined values into your models.

Setting a Parameter¶

You can define a parameter using the CLI:

bauplan parameter set --name prompt_summary --type str --value "Write a concise, incisive summary of the news article." --description "Prompt passed to GPT for article summarization"

This adds the following entry to your bauplan_project.yml:

parameters:
    prompt_summary:
    type: str
    default: Write a concise, incisive summary of the news article.
    description: Prompt passed to GPT for article summarization

Setting from a File¶

For longer values like LLM prompts, SQL templates, or configuration blobs, you can load a parameter from a file:

bauplan parameter set --name prompt_template --type str --file prompts/summary.txt

Listing and Removing Parameters¶

To list all defined parameters:

bauplan parameter ls

To remove a parameter:

bauplan parameter rm --name prompt_summary

Using Parameters in Code¶

Once defined, parameters become inputs to your models. Declare them using bauplan.Parameter(...), and Bauplan will inject their values at runtime:

@bauplan.model()
@bauplan.python('3.11')
def example_model(
    data=bauplan.Parameter("your_table"),
    my_param=bauplan.Parameter("param_name")
):
    print(my_param)  # Treated as a regular Python variable
    ...
    return data

Example: Summarize Text Using GPT¶

Suppose you have a step in your pipeline that summarizes news articles using an LLM. The OpenAI API key is stored as a secret, and the summary prompt is a configurable string parameter.

@bauplan.model(internet_access=True)  # Enables API access
@bauplan.python('3.11', pip={'openai': '1.57.2', 'pandas': '2.2.2'})
def article_summaries(
    articles=bauplan.Model("article_table"),               # Input table
    openai_key=bauplan.Parameter('openai_api_key'),        # Encrypted secret
    summary_prompt=bauplan.Parameter("prompt_summary")     # Prompt string
):
    import openai
    import pandas as pd

    openai.api_key = openai_key  # Decrypted and injected at runtime

    df = articles.to_pandas()
    summaries = []

    for _, row in df.iterrows():
        response = openai.ChatCompletion.create(
            model="gpt-4",
            messages=[
                {"role": "system", "content": summary_prompt},
                {"role": "user", "content": row["article_content"]}  # Assumes a 'article_content' column
            ]
        )
        summaries.append(response["choices"][0]["message"]["content"])

    df["llm_summary"] = summaries
    return df  # Return enriched table

What This Model Does¶

• Reads articles from a table.

• Uses a configurable prompt and a secret API key to call the OpenAI API.

• Appends a new column llm_summary with the model’s output.

• Returns the enriched table as a result.

The combination of parameters and secrets makes this pipeline flexible, secure, and maintainable — all without needing to hardcode logic or credentials.