Superagentic AI Open-Sources SuperOptiX Agent Optimization Engine

Shashikant Jagtap

Superagentic AI is open sourcing SuperOptiX. This is a major milestone in our journey and a practical step for teams building production agentic systems in a fast-moving ecosystem.

Where SuperOptiX started

SuperOptiX first launched in July 2025 as a DSPy-first framework with additional agentic capabilities for real-world execution. The early goal was straightforward: help builders ship useful agents faster while reducing repetitive glue code around prompts, model setup, tools, and orchestration.

How it evolved

As more teams adopted multiple frameworks, the same pain points kept appearing: duplicated runtime wiring, framework-specific boilerplate, optimization logic embedded directly in business pipelines, and brittle provider integrations.

SuperOptiX evolved from a framework layer into a shared optimization engine that keeps each target framework native while centralizing the hard parts.

SuperOptiX as a cross-framework optimization engine

SuperOptiX now supports multiple frameworks while preserving each framework’s native style:

  • DSPy
  •  Pydantic AI
  • Google ADK
  • OpenAI Agents SDK
  • DeepAgents
  • CrewAI
  • Claude Agent SDK
  • Microsoft Agent Framework (legacy support)

Why GEPA matters

GEPA is the backbone of SuperOptiX optimization:

  • Base compile produces a clean, runnable pipeline.
  • --optimize enables the optimization and evaluation lifecycle.
  • Optimization behavior remains inspectable and tunable without bloating default runtime code.

Why open source now

Agentic AI is advancing quickly: richer tool use, longer-context reasoning, multi-step planning, and deeper enterprise integration. Closed and rigid stacks cannot keep pace with this rate of change. Open ecosystems, composable tooling, and transparent generated code are now practical requirements.

What open source SuperOptiX enables

  • Framework-native pipelines with less boilerplate.
  • An explicit and modern GEPA-powered optimization lifecycle.
  • Early support paths for research patterns such as RLM (experimental).
  • Connector-driven workflows, including modern integration paths such as StackOne.

What comes next

  • Deeper GEPA-first optimization workflows
  • More connector-first agent capabilities
  • Continued RLM experimentation
  • Minimal, readable, framework-native generated pipelines

Thank you

Thank you to everyone who tested early builds, filed issues, and pushed us toward cleaner framework-native behavior. Your feedback directly shaped this release.

SuperOptiX is now open source, and this is just the beginning.