DocsModels

Models

Use the latest AI models from OpenAI, Google, and Anthropic with a single interface.

Download

Quick Start (60 Seconds)

Easiest Way: Use Managed Keys

No API key setup required! Authenticate once and get 100K free tokens to start.

Terminal

$co auth

main.py

from connectonion import Agent

# Add co/ prefix - that's it!
agent = Agent("assistant", model="co/gpt-5")
response = agent.input("Explain quantum computing")

Python REPL

Interactive

Quantum computing harnesses quantum mechanical phenomena...

⭐ Bonus: Star our repo for an additional 100K tokens!

Alternative: Bring Your Own Keys

For production or high-volume usage, use your own API keys for direct billing.

Terminal

$export OPENAI_API_KEY="sk-..."

main.py

from connectonion import Agent

# Use model names directly
agent = Agent("assistant", model="gpt-5")
response = agent.input("Explain quantum computing")

Python REPL

Interactive

Quantum computing harnesses quantum mechanical phenomena...

Default Models

When you don't specify a model, ConnectOnion uses these optimized defaults:

Agent Default

For Agent() class

main.py

# Default: co/gemini-2.5-pro
agent = Agent("assistant")  # Uses co/gemini-2.5-pro

# Same as:
agent = Agent("assistant", model="co/gemini-2.5-pro")

Best for agentic tasks with tool calling and complex reasoning

llm_do Default

For llm_do() function

main.py

from connectonion import llm_do

# Default: co/gemini-2.5-flash
result = llm_do("Summarize this text...")

# Same as:
result = llm_do("...", model="co/gemini-2.5-flash")

Fast and cost-effective for simple LLM calls

Note: Default models require authentication with co auth. To use your own API keys without the co/ prefix, set GEMINI_API_KEY environment variable.

Available Models

GPT-5 Series

gpt-5Best for coding and agentic tasks across domains

gpt-5-miniFaster, cost-efficient version for well-defined tasks

gpt-5-nanoFastest, most cost-efficient version

Reasoning Models

o4-miniOpenAI's newest reasoning model

Model Selection Guide

Best for Coding

main.py

agent = Agent("coder", model="gpt-5")
# Alternative: claude-sonnet-4-5

Fast Responses

main.py

agent = Agent("quick", model="gpt-5-nano")
# Alternative: gemini-2.5-flash

Cost-Optimized

main.py

agent = Agent("budget", model="gpt-5-nano")
# Alternative: gemini-2.5-flash-lite

Multimodal

main.py

agent = Agent("vision", model="gemini-2.5-pro")
# Supports: audio, video, images, PDF

Two Ways to Use Models

Option 1: Managed Keys

Recommended for getting started

Terminal

$co auth

main.py

# Use any model with co/ prefix
agent = Agent("assistant", model="co/gpt-5")
agent = Agent("assistant", model="co/gemini-2.5-pro")
agent = Agent("assistant", model="co/claude-opus-4-5")

✓ 100K free tokens to start

✓ Access to all providers

✓ No API key management

Option 2: Your Own Keys

For production or high-volume usage

Terminal

$export OPENAI_API_KEY="sk-..."

$export GEMINI_API_KEY="AIza..."

$export ANTHROPIC_API_KEY="sk-ant-..."

main.py

# Use models without co/ prefix
agent = Agent("assistant", model="gpt-5")
agent = Agent("assistant", model="gemini-2.5-pro")
agent = Agent("assistant", model="claude-opus-4-5")

✓ Production deployments

✓ Direct billing

✓ Existing infrastructure

Smart Model Selection

Automatically select the best model based on your needs:

main.py

def select_model(task_type: str, speed_priority: bool = False) -> str:
    """Select optimal model based on requirements."""
    
    if speed_priority:
        return {
            "code": "gpt-5-mini",
            "chat": "gpt-5-nano",
            "analysis": "gemini-2.5-flash"
        }.get(task_type, "gpt-5-nano")
    else:
        return {
            "code": "gpt-5",
            "reasoning": "o4-mini",
            "analysis": "claude-opus-4-5"
        }.get(task_type, "gpt-5")

# Use appropriate model
model = select_model("code", speed_priority=False)
agent = Agent("coder", model=model)

Python REPL

Interactive

Selected model: gpt-5

Fallback Chain

Try multiple models if one fails:

main.py

def create_agent_with_fallback(name: str):
    """Try multiple models if one fails."""
    
    model_chain = [
        "gpt-5",           # Best overall
        "claude-opus-4-5", # Strong alternative
        "gemini-2.5-pro",  # Multimodal option
        "gpt-5-mini"       # Faster fallback
    ]
    
    for model in model_chain:
        try:
            return Agent(name, model=model)
        except Exception as e:
            print(f"Failed with {model}: {e}")
            continue
    
    raise Exception("No models available")

# Will use best available model
agent = create_agent_with_fallback("assistant")

Python REPL

Interactive

Using model: gpt-5

Model Comparison

Model	Provider	Context	Strengths
`gpt-5`	OpenAI	200K	Best for coding & agentic tasks
`gemini-3-pro-preview`	Google	1M	State-of-the-art reasoning
`gemini-2.5-pro`	Google	2M	Multimodal, huge context
`claude-opus-4-5`	Anthropic	200K	Most capable Claude

Key Benefits

• Get started in 60 seconds with managed keys (co auth)
• 100K free tokens + bonus credits for starring our repo
• Same code works everywhere - just change the model name or add/remove co/ prefix
• Tool support identical across all models and providers
• Easy transition from managed keys to your own keys when ready for production

Logging

LLM Function