Prompt Bottle

A powerful prompt template engine built upon Jinja

Features ✨

• 🎯 Jinja-Powered Templates: Leverage the full power of Jinja2 templating for dynamic prompt generation
• 🎭 Role-Based Messaging: Structure conversations with system, user, assistant, and tool roles
• 🤖 Pydantic-AI Compatible: Returns structured pydantic-ai ModelMessage objects for seamless integration
• 🔄 Format Conversion: Convert to OpenAI chat API (or other APIs) through pydantic-ai's message mapping

Installation 📦

pip install prompt-bottle

Quick Start 🚀

Basic Usage

from prompt_bottle import render

# Simple template
template = """
You are a helpful assistant.
<div role="user">{{ user_message }}</div>
"""

messages = render(template, user_message="Hello, world!")
print(messages)

Core Concepts 📚

Template Example

You are a helpful assistant.
{{ system_instructions }}

{% for item in conversation %}
<div role="user">{{ item.message }}</div>
<div role="assistant">
  <think>{{ item.reasoning }}</think>
  {{ item.response }}
</div>
{% endfor %}

<div role="user">
  <Instruction>
  Now you are required to answer the query based on the context.
  Your output must be quoted in <Answer></Answer> tags.
  </Instruction>
  <Context>
  {{ context }}
  </Context>
  <Query>
  {{ query }}
  </Query>
</div>

💡 Check out example.py and example.jinja for a comprehensive example with all features!

Roles

Prompt Bottle supports four main roles:

• system: System instructions and configuration. Default of raw text.
• user: User messages and queries
• assistant: AI assistant responses
• tool: Tool execution results

Response Types

Assistant responses can include multiple content tags:

• <text>: Regular text content. Default of raw text.
• <tool_call>: Function/tool invocations
• <think>: Reasoning and thought processes

Request Types

TODO: Will support multimodal input in the future.

API Reference 📖

render(template: str, **kwargs) -> list[ModelMessage]

Renders a Jinja template with the provided variables and returns structured messages.

Parameters:

• template: Jinja template string
• **kwargs: Template variables

Returns: List of pydantic-ai ModelMessage objects (ModelRequest/ModelResponse)

to_openai_chat(messages: list[ModelMessage], **model_kwargs) -> list[dict]

Converts structured messages to OpenAI chat completion format.

Parameters:

• messages: List of ModelMessage objects from render()
• **model_kwargs: OpenAI model configuration (default model: gpt-4o)

Returns: List of OpenAI-formatted message dictionaries

Contributing 🤝

Contributions are welcome! Please feel free to submit a Pull Request.

License 📄

This project is licensed under the MIT License - see the LICENSE file for details.

Links 🔗

velin: Vue.js based prompt template engine