Skip to content
SuperOptiX Logo

Introduction to SuperOptiX

The world's first universal agent optimization framework

🔍 What is SuperOptiX?

SuperOptiX is the world's first universal agent optimization framework that brings end-to-end optimization to AI agents across 6 major frameworks. Build, evaluate, and optimize agents from any framework with a unified workflow powered by GEPA (Genetic-Pareto Algorithm).

Unlike most frameworks that bolt on evals and monitoring as an afterthought, SuperOptiX makes evaluation, optimization, and guardrails core to the development lifecycle. Whether you're using DSPy, OpenAI SDK, CrewAI, Google ADK, Microsoft, or DeepAgents, SuperOptiX helps you ship reliable agents.

SuperOptiX brings together declarative agent specification, universal GEPA optimization, built-in evaluation, multi-framework support, and multi-agent orchestration, all grounded in the principles of test-driven development and context engineering.

What's New

  • 🧪 RLM Support (Experimental): Available for early adoption and testing. Unified sandbox support is coming soon. See RLM guide.
  • 🗂️ Connector Integrations: StackOne connector support for cross-framework tool-driven agents. See StackOne integration.
  • ⚙️ Minimal + Optimize Flow: Default compile generates minimal framework-native pipelines, with expanded optimization flow enabled via --optimize.

Core Philosophy

Unlike frameworks that lock you into a single approach, SuperOptiX gives you freedom of choice with consistency of experience. Build agents in DSPy, OpenAI SDK, CrewAI, Google ADK, Microsoft Agent Framework, or DeepAgents, then optimize them all with the same powerful GEPA optimizer.

Key Principle

Declarative by Design. Universally Optimized. Framework-Agnostic.

With its native DSL (SuperSpec), GEPA optimization engine, support for major frameworks, and full-stack abstractions, SuperOptiX empowers you to build reliable, adaptive, and intelligent agentic systems without vendor lock-in.

🔮 Future-Proof by Design

SuperOptiX provides true extensibility across three critical abstraction layers:

🔬 Framework Layer

Swap AI frameworks on demand

Currently supported: DSPy, OpenAI SDK, CrewAI, Google ADK, Microsoft Agent Framework, DeepAgents

Future ready: New framework? Add an adapter. Your agents keep working.

target_framework: any_framework

🔌 Protocol Layer

Swap communication protocols

Currently supported: MCP (Model Context Protocol)

Future ready: A2A, custom protocols. Protocol-first design means seamless adoption.

protocol: mcp | a2a | custom

🧬 Optimizer Layer

Swap optimization engines

Currently supported: GEPA, DSPy optimizers (BootstrapFewShot, MIPRO, MIPROv2)

Future ready: New optimizer research? Plugin system makes integration trivial.

optimizer: gepa | mipro | custom

💎 Abstraction Without Lock-In

Your agent context stays constant. Frameworks evolve. Protocols emerge. Optimizers improve.

SuperOptiX adapts without breaking your agents.

Built for today. Ready for tomorrow. Future-proof forever.

🏗️ Full-Stack Agentic Optimization

SuperOptiX provides end-to-end optimization across every layer of the agentic stack. While other frameworks optimize prompts in isolation, SuperOptiX takes a holistic approach with universal GEPA optimization that works independently of any specific framework:

💬 Prompts & Instructions

Evolutionary optimization of:

  • System prompts
  • Agent instructions
  • Reasoning patterns
  • Task descriptions

📚 RAG & Retrieval

Context optimization across:

  • Retrieval queries
  • Chunking strategies
  • Embedding selection
  • Relevance scoring

🛠️ Tools & Functions

MCP protocol optimization for:

  • Tool selection
  • Parameter tuning
  • Execution strategies
  • Error handling

🧠 Memory & Context

Intelligent optimization of:

  • Memory retrieval
  • Context windows
  • Storage strategies
  • Attention patterns

⚡ Universal Optimization Engine

Framework-agnostic optimization means GEPA works the same whether you're using DSPy, CrewAI, OpenAI SDK, or any other framework.

No framework-specific tricks. No vendor lock-in. Just pure, portable optimization across your entire agentic stack.

Optimize everything. Switch anything. Break nothing.

⚡ Core Features

🎯 Evaluation-First Architecture

RSpec-style BDD specifications with built-in testing and validation from day one. Define behavior scenarios, run automated evaluations, track metrics across optimization iterations.

🧬 GEPA Universal Optimizer

Automatic optimization across all 6 major agent frameworks with proven results. Achieve 37.5% → 80% improvements with just 3-10 training scenarios.

🔬 Multi-Framework Support

Works with DSPy, OpenAI SDK, CrewAI, Google ADK, Microsoft, and DeepAgents. One workflow, six frameworks, unlimited possibilities.

📝 SuperSpec DSL

Declarative language for agent specifications with Kubernetes-style versioning. Write once, deploy to any framework.

🧠 Context Engineering

Systematic approach to delivering optimal information and tools to agents. Includes RAG, MCP protocol, and memory optimization.

🏗️ Framework Architecture

Progressive capabilities with built-in safety, evaluation, and optimization workflows.

🌟 How SuperOptiX Differs from Other Frameworks

Feature Other Frameworks SuperOptiX
🎯 Evaluation-First Add evaluation as an afterthought Evaluation built into core development cycle
RSpec-Style BDD Development Manual prompt engineering Behavior-driven specifications with automated testing
🧬 Universal GEPA Optimizer Manual optimization or framework-specific tools One optimizer for all frameworks
🔬 Multi-Framework Freedom Locked into one approach Choose from multiple frameworks, optimize with one tool
🚀 Deployment-Ready Basic deployment capabilities Built-in observability, guardrails, and enterprise features

Multi-Framework Support: Choose Your Tool, Keep Your Workflow

SuperOptiX is the world's first framework-agnostic agent optimizer. Build agents in any of the 6 major frameworks, then optimize them all with the same powerful GEPA engine.

🔬 Supported Frameworks

Framework Variables Local Models Best For Status
🔬 DSPy 10+ Ollama Complex reasoning, research Proven: 37.5% → 80%
🤖 OpenAI SDK 1 Ollama Simple & fast Proven: 100% pass rate
👥 CrewAI 5 Ollama Multi-agent teams Proven: 100% pass rate
🔮 Google ADK 1 ☁️ Cloud Gemini native, free access Ready
🏢 Microsoft 1 Ollama Enterprise Azure Ready
🌊 DeepAgents 1 Ollama Complex planning Ready

One Workflow, All Frameworks

# Same workflow regardless of framework!
super agent compile my_agent --framework [dspy|openai|crewai|google-adk|microsoft|deepagents]
super agent evaluate my_agent
super agent optimize my_agent --auto medium  # GEPA works on ALL frameworks
super agent run my_agent

Learn More

See our Multi-Framework Support Guide for detailed comparisons and examples.

Universal Workflow

SuperOptiX provides a consistent workflow across all frameworks:

  1. Specify: Define your agent using SuperSpec DSL
  2. Compile: Generate framework-specific code
  3. Evaluate: Test with RSpec-style BDD scenarios
  4. Optimize: Improve with GEPA
  5. Deploy: Run
# Initialize project
super init my_project
cd my_project

# Pull a pre-built agent
super agent pull sentiment_analyzer

# Compile for your chosen framework
super agent compile sentiment_analyzer

# Evaluate performance
super agent evaluate sentiment_analyzer

# Optimize with GEPA
super agent optimize sentiment_analyzer --auto medium

# Run
super agent run sentiment_analyzer

Architecture Overview

SuperOptiX consists of several key components:

  • SuperSpec DSL: Declarative specification language
  • GEPA Optimizer: Universal optimization engine
  • Framework Adapters: Support for 6 major frameworks
  • Evaluation Engine: RSpec-style BDD testing framework
  • Memory System: Multi-layered memory management
  • RAG Integration: Knowledge retrieval across frameworks
  • Orchestra: Multi-agent coordination
  • Observability: Built-in monitoring and tracing

📈 Proven Results

GEPA has delivered dramatic improvements across all tested frameworks:

Framework Baseline After GEPA Improvement
🔬 DSPy 37.5% 80.0% +42.5 pts 🏆
🤖 OpenAI SDK 100% 100% Maintained
👥 CrewAI 75% 100% +25 pts ⭐
🔮 Google ADK Ready for optimization
🏢 Microsoft Ready for optimization
🌊 DeepAgents Ready for optimization

Sample Efficient

GEPA achieves these improvements with just 3-10 training scenarios, while traditional methods require hundreds of examples.

🚀 Getting Started

Ready to build your first optimized agent?

1️⃣

Quick Start Guide

Build your first agent in 10 minutes

2️⃣

Installation

Set up your development environment

3️⃣

LLM Setup

Configure your language models

4️⃣

SuperSpec Guide

Learn the specification language

5️⃣

GEPA Optimization

Master universal optimization

🏗️ Capability Coverage

SuperOptiX focuses on practical capability coverage:

  • Framework support (DSPy, OpenAI SDK, CrewAI, Google ADK, Microsoft, DeepAgents, Pydantic AI)
  • SuperSpec-based agent definitions
  • Built-in evaluation and optimization loops
  • Observability and orchestration support

Community and Support

🎯 Next Steps

🚀

Quick Start

Build your first agent in 10 minutes

📚

Guides

In-depth tutorials and guides

⚙️

API Reference

Complete API documentation

💡

Examples

Real-world agent examples

🚀 Ready to Build Optimized AI Agents?

Start with quick start and build your first agent today.