fireprompt 0.1.9

Turn Python functions into LLM prompts

🔥 fireprompt

A lightweight Python library that turns your functions into LLM prompts. Write Jinja2 templates, get type-safe Pydantic responses, and connect to 100+ providers via LiteLLM.

FirePrompt is a framework, not a prompt library. You define your own functions (analyze_sentiment, extract_data, generate_code - whatever you need). FirePrompt handles the LLM integration, type validation, and templating.

Links:

• Quick Start
• Setup & Authentication
• Configuration
• Callbacks
• Features
• License

Quick Start

pip install fireprompt

Setup & Authentication

FirePrompt uses LiteLLM to access 100+ LLM providers (OpenAI, Anthropic, Google, Azure, AWS Bedrock, and more).

To get started: Set up your provider's API key as an environment variable. Each provider has different requirements.

📚 Follow the setup guide for your provider: 👉 LiteLLM Provider Documentation

Example for OpenAI:

export OPENAI_API_KEY="sk-..."

Then in your code:

model=LLM(name="openai/gpt-4o")

How It Works

1. Write a Python function - Define what you want (e.g., research_topic, analyze_sentiment, generate_code)
2. Add the @prompt decorator - Tell FirePrompt which LLM to use
3. Write your prompt in the docstring - Use Jinja2 templates to inject variables
4. Call it like a normal function - FirePrompt handles the LLM call and returns typed results

Example: Here's a custom prompt function you might create:

from fireprompt import prompt, LLM
from pydantic import BaseModel

class ResearchResult(BaseModel):
    text: str

@prompt(
    model=LLM(
        name="openai/gpt-4o"
    )
)
def research_topic(topic: str, level: str = "basic") -> ResearchResult:
    """
    - role: system
      content: You are a helpful research assistant.

    - role: user
      content: >
        Analyze {{ topic }} at a {{ level }} level.
        Provide a clear and informative overview.
    """
    ...

result = research_topic(topic="Quantum Computing")
print(result.text)

Output:

Quantum computing is a revolutionary computing paradigm that leverages quantum mechanics
principles to process information. Unlike classical computers that use bits (0 or 1),
quantum computers use qubits that can exist in superposition, representing both states
simultaneously. This enables quantum computers to solve certain complex problems
exponentially faster than classical computers, particularly in cryptography, optimization,
and molecular simulation.

Configuration

Custom Parameters

Use LLMConfig for universal parameters across all providers:

from fireprompt import prompt, LLM, LLMConfig
from pydantic import BaseModel

class Summary(BaseModel):
    text: str

@prompt(
    model=LLM(
        name="anthropic/claude-3-5-sonnet-20241022",
        config=LLMConfig(
            temperature=0.3,  # Lower = more focused (0-2)
            max_tokens=1000   # Max response length
        )
    )
)
def summarize(text: str, style: str = "professional", keywords: list[str] = []) -> Summary:
    """
    - role: system
      content: You are an expert text summarizer.

    - role: user
      content: |
        Summarize this text in {{ style | upper }} style.

        {% if keywords %}
        Focus on these keywords:
        {% for keyword in keywords %}
        - {{ keyword | title }}
        {% endfor %}
        {% endif %}

        Text:
        {{ text }}
    """
    ...

with open("article.txt") as f:
    result = summarize(
        text=f.read(),
        style="casual",
        keywords=[
            "AI",
            "Machine Learning",
            "Innovation"
        ]
    )

print(result.text)

Provider-Specific Parameters

Use extra for provider-specific options:

from fireprompt import prompt, LLM, LLMConfig
from pydantic import BaseModel

class Analysis(BaseModel):
    summary: str
    key_points: list[str]

@prompt(
    model=LLM(
        name="openai/gpt-4o",
        config=LLMConfig(
            temperature=0.7,
            extra={
                "frequency_penalty": 0.5,  # OpenAI-specific
                "presence_penalty": 0.3
            }
        )
    )
)
def analyze(text: str) -> Analysis:
    """
    - role: user
      content: Analyze this text: {{ text }}
    """
    ...

result = analyze("Your text here...")
print(result)

Presets

Pre-configured settings for common use cases:

from fireprompt import prompt, LLM, LLMConfigPreset
from pydantic import BaseModel

class BlogPost(BaseModel):
    title: str
    content: str

@prompt(
    model=LLM(
        name="gemini/gemini-1.5-pro",
        config=LLMConfigPreset.CREATIVE
    )
)
def write_blog(topic: str, sections: list[str] = []) -> BlogPost:
    """
    - role: user
      content: |
        Write an engaging blog post about {{ topic | title }}.

        {% if sections | length > 0 %}
        Include these sections:
        {% for section in sections %}
        {{ loop.index }}. {{ section }}
        {% endfor %}
        {% else %}
        Create your own structure.
        {% endif %}
    """
    ...

result = write_blog("Generative AI", sections=["Introduction", "Key Benefits", "Use Cases"])
print(result)

Preset	Temperature	Top P	Best For
DEFAULT	0.7	-	Balanced responses
CREATIVE	0.9	0.95	Writing, brainstorming
PRECISE	0.3	0.9	Technical analysis
DETERMINISTIC	0.0	1.0	Reproducible outputs

Callbacks

Post-process responses with on_complete:

from fireprompt import prompt, LLM
from pydantic import BaseModel

class Analysis(BaseModel):
    sentiment: str
    score: float

def log_result(result: Analysis) -> Analysis:
    print(f"✓ {result.sentiment} ({result.score})")
    return result

@prompt(
    on_complete=log_result,
    model=LLM(
        name="openai/gpt-4o-mini"
    )
)
def analyze_sentiment(text: str) -> Analysis:
    """
    - role: user
      content: >
        Analyze sentiment: {{ text }}
        Return sentiment (positive/negative/neutral) and score (0-1).
    """
    ...

result = analyze_sentiment("I love this!")  # ✓ positive (0.95)
print(result)

Debug Logging

import fireprompt

fireprompt.enable_logging()   # Enable debug mode
fireprompt.disable_logging()  # Disable debug mode

Features

Feature	Description
🎯 Type-Safe	Automatic Pydantic validation for structured outputs
⚡ Async First	Auto-detects and handles sync/async functions
🎨 Jinja2 Templates	Full template support with variables, filters, loops, conditionals
🔧 Flexible Config	Universal parameters + provider-specific options via extra
📦 Presets	Pre-configured settings for common use cases
🪝 Callbacks	Post-process responses with on_complete
🔍 Debug Mode	Built-in logging for development
🚀 Multi-Provider	100+ LLMs via LiteLLM - OpenAI, Anthropic (Claude), Google (Gemini), Azure, AWS Bedrock, and more

License

MIT License - see LICENSE for details.