🎯 What is OpenAI Agents SDK?
A lightweight yet powerful framework for building multi-agent workflows
Provider-Agnostic
Works with OpenAI, Ollama, and 100+ LLMs
Multi-Agent
Built-in handoffs for agent collaboration
Guardrails
Input/output validation built-in
Sessions
Automatic conversation history management
Tracing
Built-in execution tracking
Works with Ollama!
Unlike DeepAgents, no function-calling limitations
Perfect for simple to moderate complexity tasks with local models!
📦 Installation
uv tool install superoptix --with "superoptix[frameworks-openai]"
Includes:
- • openai-agents 0.20.x
- • openai SDK (latest)
- • SuperOptiX core with GEPA 0.1.4
Requirements:
- • Python 3.11+
- • Git (for DSPy dependency)
🚀 Quick Start
Get started with OpenAI Agents SDK in minutes - works with FREE Ollama!
Pull the Demo Agent
super agent pull assistant_openai
Configure Model - Uses Ollama by Default!
Uses Ollama by Default! (FREE, no API keys needed!)
# The assistant_openai agent defaults to Ollama llama3.1:8b # Just install Ollama and run: brew install ollama # macOS ollama pull llama3.1:8b super agent run assistant_openai --goal "Hello!"
Run the Workflow
# Compile super agent compile assistant_openai --framework openai # Evaluate super agent evaluate assistant_openai # Optimize with GEPA super agent compile assistant_openai --framework openai --optimize super agent optimize assistant_openai --auto light --framework openai --reflection-lm ollama:llama3.1:8b # Run super agent run assistant_openai --goal "What is Python?"
🔄 Model Configuration
Ollama (Local)
language_model: location: local provider: ollama model: ollama:llama3.1:8b temperature: 0.7 api_base: http://localhost:11434
Advantages:
- Free inference
- Privacy (data stays local)
- Fast development iteration
- Good baseline performance
Cloud Models
# OpenAI GPT-4 language_model: location: cloud provider: openai model: openai:gpt-4o # Set: export OPENAI_API_KEY="sk-..." # Anthropic Claude language_model: location: cloud provider: anthropic model: anthropic:claude-sonnet-4-20250514 # Set: export ANTHROPIC_API_KEY="sk-ant-..."
🎯 What GEPA Optimizes
Optimization Target
instructions
The agent's system prompt - GEPA generates variations, tests each, and selects the best performer
Example Transformation:
Before: "Helpful AI Assistant\nGoal: Provide clear responses"
After GEPA: "You are a Helpful AI Assistant.\n\nWhen answering questions:\n1. Read carefully\n2. Provide accurate info\n3. Use clear language\n4. Be concise but complete"
Optimization Command
super agent optimize assistant_openai \ --auto medium \ --framework openai \ --reflection-lm ollama:llama3.1:8b
Note: Requires --reflection-lm parameter for Universal GEPA optimization
⚙️ Advanced Features
🔧 Tools Support
tools: enabled: true specific_tools: - name: search_database description: Search knowledge base - name: send_email description: Send email to user
🤝 Multi-Agent Handoffs
handoffs:
- name: specialist_agent
description: Technical questions
instructions: You are a technical specialist
- name: billing_agent
description: Billing issues🛡️ Guardrails
guardrails: input_guardrails: - type: content_safety params: block_harmful: true output_guardrails: - type: pii_filter params: redact_emails: true
OpenAI SDK vs Other Frameworks
| Feature | DSPy | DeepAgents | OpenAI SDK |
|---|---|---|---|
| Ollama Support | Full | ❌ Blocked | Perfect |
| API Complexity | Medium | High | Low |
| Multi-Agent | Manual | Subagents | Handoffs |
| GEPA Optimization | All signatures | system_prompt | instructions |
| Best For | Prompt optimization | Complex planning | Simple tasks |
Perfect for:
- Simple to moderate complexity tasks
- Local development with Ollama
- Quick prototyping
- Clean, minimal agent design
- When you want high baseline performance
❌ Not ideal for:
- Complex multi-step planning (use DeepAgents)
- Maximum prompt optimization (use DSPy)
- Need filesystem context management (use DeepAgents)