The Core of SuperOptiX

Full-Stack Agent Optimization

SuperOptiX is the only framework that optimizes across every layer of the agentic stack. From prompts to RAG, tools to memory, protocols to datasets — one platform, complete optimization.

Optimize Every Layer

Traditional frameworks optimize prompts only. SuperOptiX optimizes the entire agentic stack.

Featured
🧬

GEPA Universal Optimizer

The world's first framework-agnostic optimizer. Works across DSPy, OpenAI, CrewAI, Google ADK, Microsoft, DeepAgents, and Pydantic AI.

Works with just 3-10 training examples
Reflective prompt evolution
Domain-specific textual feedback
Learn More →
💬

Prompt Optimization

Evolutionary optimization of system prompts, instructions, and reasoning patterns using GEPA's reflective analysis.

System prompt optimization
Agent instruction tuning
Reasoning pattern enhancement
Task description refinement
View in Docs
📚

RAG Optimization

Optimize retrieval queries, chunking strategies, and context selection for maximum relevance and accuracy.

Retrieval query optimization
Chunking strategy tuning
Embedding selection
Context window optimization
View in Docs
🛠️

Tool Optimization

MCP-based tool optimization for better parameter selection, execution strategies, and tool coordination.

Tool selection optimization
Parameter tuning
Execution strategy refinement
Multi-tool coordination
View in Docs
🧠

Memory Optimization

GEPA-optimized memory selection to fit within LLM context windows, improving quality while reducing token usage.

Up to 60% token reduction
Relevant memory selection
Context window optimization
Long-term memory management
View in Docs
🔌

Protocol Optimization (MCP)

Optimize Model Context Protocol integration for enhanced tool communication and agent interoperability.

MCP server optimization
Tool discovery tuning
Protocol efficiency
A2A preparation
View in Docs
🔬

DSPy Optimizers

Advanced DSPy optimizers including BootstrapFewShot, MIPROv2, and SIMBA for specialized optimization scenarios.

BootstrapFewShot for few-shot learning
MIPROv2 for instruction optimization
SIMBA for complex pipelines
Custom optimizer support
View in Docs
📊

Dataset-Driven Optimization

Import and leverage datasets from CSV, JSON, Parquet, and HuggingFace for data-driven optimization.

Multiple format support
HuggingFace integration
Automatic conversion
Evaluation datasets
View in Docs
Proven Results

Real Optimization Results

GEPA delivers proven performance improvements across all supported frameworks.

FrameworkAgentBaselineAfter GEPAImprovement
🔬 DSPySentiment Analyzer
37.5%
80.0%
+42.5 pts 🏆
👥 CrewAIResearch Crew
75%
100%
+25 pts ⭐
🤖 OpenAI SDKAI Assistant
100%
100%
Maintained
🔍 Google ADKGemini Assistant
Ready
Available
💠 MicrosoftEnterprise Agent
Ready
Available
🐍 Pydantic AIType-Safe Agent
Ready
Available

One Command, All Frameworks

The same optimization workflow works across every supported framework.

# Step 1: Initialize and pull agent
$ super init my_project && cd my_project
$ super agent pull sentiment_analyzer
# Step 2: Compile and evaluate baseline
$ super agent compile sentiment_analyzer
$ super agent evaluate sentiment_analyzer
📊 Baseline: 37.5% accuracy
# Step 3: Optimize with GEPA (works for ANY framework!)
$ super agent optimize sentiment_analyzer --auto medium
🧬 GEPA optimizing... ✓ Complete!
# Step 4: Evaluate optimized version
$ super agent evaluate sentiment_analyzer
📊 After GEPA: 80.0% accuracy (+42.5 pts!) 🏆

Ready to Optimize Your Agents?

Start with GEPA and experience the power of universal agent optimization.