​

🧾 Prompt Database

Prompt Database is your centralized, versioned store for managing agent prompts — without needing to redeploy code or edit files in a repo.

It allows you to:

• Create and version prompts in the UI
• Dynamically fetch them at runtime
• Templatize with variables
• Rapidly iterate on prompt design
• Swap/Adjust prompts without touching your codebase

​

🚀 Why Prompt DB?

Improving an AI agent often means adjusting its prompts. But doing that through your codebase slows everything down.

With our Prompt DB, you can:

• Change a prompt in the dashboard → rerun your agent instantly without redeploying
• Test prompt performance across Sessions or Mass Sims
• Compare and experiment with different versions of prompts without redeploying anything

📷

​

✍️ How to Create One

​

Step 1: In the UI

• Click a project and then an agent
• On the agent page it will be directly below the agent name, so in the example directly below “Browseruse Demo”

📷

• Click Create New Prompt button.

📷

• Create a prompt name (this is what you will refer to it in the Python Package)
• Add a version description (optional)
• Copy paste your prompt from your code and change all your variables to the format {{variable_name}}.

For example, your prompt in python might look like f"summarize this page: {page_text}" then when you copy it into the Prompt Database change it to "summarize this page: {{page_text}}"

• Click Save

📷

Note: ALWAYS save a prompt if you make changes to it that you would like to keep.

​

Step 2: In Python

Wherever you want to use the prompt, you can pull it from our Prompt Database using the following code:

prompt = lai.get_prompt(
    name="summarize_page",
    label="production",  # optional, defaults to "production"
    variables={"page_text": "Stanford University is a top..."},  # optional
    cache_ttl=300  # optional (in seconds) defaults to 300
)

📘 If your prompt references {{variable_name}} but you don’t provide a value, you’ll get a warning. If you provide variables that aren’t used in the prompt, you’ll get an error

​

🏷️ Labels

Labels attach to a single prompt version and allow flexible, dynamic fetches. They’re a powerful feature that enables prompt management without code changes.

​

What Labels Do

• Version Identification: Labels point to specific prompt versions without needing IDs
• Runtime Flexibility: Switch versions without code changes
• Environment Management: Separate development, test, and production prompts
• A/B Testing: Direct different users to different versions
• Rollback Support: Quickly revert to previous versions by moving a label if a new version isn’t performing well

​

Available Labels

​

How Labels Work

• Only one version can hold a given label at a time
• If you move a label to a new version, the old version will no longer have that label
• When you create a new prompt version, the latest label automatically moves to it
• Moving the production label is a manual action (to ensure stability)
• Custom labels (like beta, experiment-a, region-specific) can be created by you for specific use cases

​

How to use labels

1. Click labels next to the version of the prompt you want to use

📷

2. Click the new label(s) you would like to move to the version

📷

3. Click Save Changes

​

🔗 Variables

Variables allow you to create dynamic, reusable prompts by inserting runtime values into your prompt templates. This is critical for creating prompts that can adapt to different inputs without creating new versions.

​

How Variables Work

1. Define variables in your prompt using double curly braces: {{variable_name}}
2. Pass variable values into the variables field in your code
3. Lucidic automatically substitutes the variable values into your prompt before sending it to the LLM

​

Formatting Variables

When adding prompts to the Prompt DB, you need to convert your code’s variable syntax to Lucidic’s template syntax:

​

Code Examples

Basic example:

In the Prompt DB UI, you would define the prompt as:

Generate {{number}} ideas about {{topic}}

In your code, you would call it like this:

prompt = lai.get_prompt(
    name="idea_generator",
    variables={
        "number": "5",
        "topic": "artificial intelligence"
    }
)

Result:

Generate 5 ideas about artificial intelligence

Another example:

In the Prompt DB UI, you would define the prompt as:

{{system_prompt}}

Answer the user's question: {{question}}

In your code, you would call it like this:

prompt = lai.get_prompt(
    name="qa_prompt",
    variables={
        "system_prompt": "You are a helpful AI assistant specialized in biology.", 
        "question": "How do cells divide?"
    }
)

Result:

You are a helpful AI assistant specialized in biology.

Answer the user’s question: How do cells divide?

​

Error Handling

• Missing variables: If your prompt references {{variable_name}} but you don’t provide a value, you’ll get a warning
• Extra variables: If you provide variables that aren’t used in the prompt, you’ll get an error

​

Best Practices

• Use descriptive variable names that indicate their purpose
• Consider setting default values for optional variables in your code
• Test your prompts with different variable values to ensure they work as expected

​

🧪 Versioning and Iteration

When you update a prompt in the UI:

• A new Prompt Version is created
• The latest label is moved to the new version
• You can optionally move the production label to the new version
• You cannot edit an old version of a prompt — if you save it, it will just create a new version

Because agents fetch prompts dynamically, the next run of your agent will use the updated prompt automatically — no deploy needed.