Home/LLM Function

One-shot LLM Calls

Make direct LLM calls with optional structured output. Supports OpenAI, Google Gemini, and Anthropic models through a unified interface.

Quick Start

main.py

from connectonion import llm_do

# OpenAI (default)
answer = llm_do("What's 2+2?")
print(answer)

# Google Gemini
answer = llm_do("What's 2+2?", model="gemini-1.5-flash")

# Anthropic Claude
answer = llm_do("What's 2+2?", model="claude-3-5-haiku-20241022")

Python REPL

Interactive

>>> answer = llm_do("What's 2+2?")

>>> print(answer)

That's it! One function for any LLM task across multiple providers.

With Structured Output

main.py

from pydantic import BaseModel

class Analysis(BaseModel):
    sentiment: str
    confidence: float
    keywords: list[str]

result = llm_do(
    "I absolutely love this product! Best purchase ever!",
    output=Analysis
)
print(result.sentiment)
print(result.confidence)
print(result.keywords)

Python REPL

Interactive

>>> print(result.sentiment)

'positive'

>>> print(result.confidence)

0.98

>>> print(result.keywords)

['love', 'best', 'ever']

Real Examples

Extract Data from Text

main.py

from pydantic import BaseModel

class Invoice(BaseModel):
    invoice_number: str
    total_amount: float
    due_date: str

invoice_text = """
Invoice #INV-2024-001
Total: $1,234.56
Due: January 15, 2024
"""

invoice = llm_do(invoice_text, output=Invoice)
print(invoice.total_amount)

Python REPL

Interactive

>>> print(invoice.total_amount)

1234.56

Use Custom Prompts

main.py

# With prompt file
summary = llm_do(
    long_article,
    system_prompt="prompts/summarizer.md"  # Loads from file
)

# With inline prompt
translation = llm_do(
    "Hello world",
    system_prompt="You are a translator. Translate to Spanish only."
)
print(translation)

Python REPL

Interactive

>>> print(translation)

Hola mundo

Quick Analysis Tool

main.py

def analyze_feedback(text: str) -> str:
    """Analyze customer feedback with structured output."""
    
    class FeedbackAnalysis(BaseModel):
        category: str  # bug, feature, praise, complaint
        priority: str  # high, medium, low
        summary: str
        action_required: bool
    
    analysis = llm_do(text, output=FeedbackAnalysis)
    
    if analysis.action_required:
        return f"🚨 {analysis.priority.upper()}: {analysis.summary}"
    return f"📝 {analysis.category}: {analysis.summary}"

# Use in an agent
from connectonion import Agent
agent = Agent("support", tools=[analyze_feedback])

Python REPL

Interactive

>>> result = analyze_feedback("The app crashes when I try to upload files!")

>>> print(result)

🚨 HIGH: Application crashes during file upload process

Supported Models

main.py

# OpenAI models
llm_do("Hello", model="gpt-4o")
llm_do("Hello", model="gpt-4o-mini")
llm_do("Hello", model="gpt-3.5-turbo")

# Google Gemini models
llm_do("Hello", model="gemini-1.5-pro")
llm_do("Hello", model="gemini-1.5-flash")

# Anthropic Claude models
llm_do("Hello", model="claude-3-5-sonnet-latest")
llm_do("Hello", model="claude-3-5-haiku-20241022")
llm_do("Hello", model="claude-3-opus-latest")

Python REPL

Interactive

>>> llm_do("Hello", model="gpt-4o")

'Hello! How can I assist you today?'

>>> llm_do("Hello", model="gemini-1.5-flash")

'Hello there! How can I help you?'

>>> llm_do("Hello", model="claude-3-5-haiku-20241022")

'Hello! How may I assist you today?'

Parameters

Parameter	Type	Default	Description
`input`	str	required	The input text/question
`output`	BaseModel	None	Pydantic model for structured output
`prompt`	str\|Path	None	System prompt (string or file path)
`model`	str	"gpt-4o-mini"	Model to use (supports OpenAI, Gemini, Claude)
`temperature`	float	0.1	Randomness (0=deterministic, 2=creative)

What You Get

One-shot execution - Single LLM round, no loops

Type safety - Full IDE autocomplete with Pydantic

Flexible prompts - Inline strings or external files

Smart defaults - Fast model, low temperature

Clean errors - Clear messages when things go wrong

Common Patterns

Data Extraction

main.py

from pydantic import BaseModel

class Person(BaseModel):
    name: str
    age: int
    occupation: str

person = llm_do("John Doe, 30, software engineer", output=Person)
print(f"Name: {person.name}")
print(f"Age: {person.age}")
print(f"Job: {person.occupation}")

Python REPL

Interactive

>>> print(f"Name: {person.name}")

Name: John Doe

>>> print(f"Age: {person.age}")

Age: 30

>>> print(f"Job: {person.occupation}")

Job: software engineer

Quick Decisions

main.py

is_urgent = llm_do("Customer says: My server is down!")
if "urgent" in is_urgent.lower():
    escalate()

Python REPL

Interactive

>>> is_urgent = llm_do("Customer says: My server is down!")

>>> print(is_urgent)

This appears to be an urgent issue that requires immediate attention.

Format Conversion

main.py

class JSONData(BaseModel):
    data: dict

json_result = llm_do("Convert to JSON: name=John age=30", output=JSONData)
print(json_result.data)

Python REPL

Interactive

>>> print(json_result.data)

{'name': 'John', 'age': 30}

Validation

main.py

def validate_input(user_text: str) -> bool:
    result = llm_do(
        f"Is this valid SQL? Reply yes/no only: {user_text}",
        temperature=0  # Maximum consistency
    )
    return result.strip().lower() == "yes"

Python REPL

Interactive

>>> validate_input("SELECT * FROM users WHERE id = 1")

True

>>> validate_input("DROP TABLE; DELETE everything")

False

Tips

1.Use low temperature (0-0.3) for consistent results
2.Provide examples in your prompt for better accuracy
3.Use Pydantic models for anything structured
4.Cache prompts in files for reusability

Comparison with Agent

Feature	llm_do()	Agent()
Purpose	One-shot calls	Multi-step workflows
Tools	No	Yes
Iterations	Always 1	Up to max_iterations
State	Stateless	Maintains history
Best for	Quick tasks	Complex automation

main.py

# Use llm_do() for simple tasks
answer = llm_do("What's the capital of France?")

# Use Agent for multi-step workflows
agent = Agent("assistant", tools=[search, calculate])
result = agent.input("Find the population and calculate density")

Python REPL

Interactive

>>> answer = llm_do("What's the capital of France?")

>>> print(answer)

The capital of France is Paris.

>>> result = agent.input("Find the population and calculate density")

>>> print(result)

I'll help you find the population and calculate the density. Let me search for the current data...

Error Handling

main.py

from connectonion import llm_do
from pydantic import ValidationError

try:
    result = llm_do("Analyze this", output=ComplexModel)
except ValidationError as e:
    print(f"Output didn't match model: {e}")
except Exception as e:
    print(f"LLM call failed: {e}")

Python REPL

Interactive

>>> try:

... result = llm_do("Analyze this", output=ComplexModel)

... except ValidationError as e:

... print(f"Output didn't match model: {e}")

Output didn't match model: 2 validation errors for ComplexModel...

Next Steps

Learn about Agents

For multi-step workflows

Explore Tools

For extending agents

See @xray

For debugging

max_iterations

Trust Parameter