OrKa is a modular, transparent AI cognition framework for orchestrating agent-based reasoning workflows using LLMs. Designed from the ground up for traceability, modularity, and introspective control, OrKa enables the construction of composable reasoning systems inspired by cognitive science — without sacrificing the power of modern foundation models.

💡 Built for AGI researchers, explainability advocates, and developers who need more than just chains and wrappers.

OrKa is a lightweight SDK and orchestration runtime for defining cognitive pipelines using YAML files and pluggable AI agents.

It shifts AI orchestration from monolithic prompting or hard-coded chains to a distributed cognition model, where each step is handled by a distinct reasoning agent: classifiers, validators, searchers and more.

Every step is logged, inspectable, and overrideable. OrKa doesn't hide the process — it lets you observe cognition as it unfolds.

• Orchestrators manage control flow between agents.
• Agents are stateless modules that perform tasks like:
  • Binary decision making
  • Classification
  • Web search
  • Validation
  • Routing (conditional branching)
• Redis Streams or (soon) Kafka enable async message passing and logging.
• YAML configuration defines the entire reasoning graph declaratively.

✅ Pluggable agent system
✅ LLM-backed binary/classification/chain-of-thought agents
✅ Google/DuckDuckGo search fallback
✅ Full Redis-based trace logging
✅ Kafka-ready memory layer (planned)
✅ Compatible with local or remote LLMs (via LiteLLM)
✅ Roadmap for visual orchestration via Tiamat-GenAI

Unlike LangChain, CrewAI, or Flowise, OrKa is not just a wrapper for LLM APIs. It is:

• 🧱 Modular by default — Agents are black-box-optional, testable, and swappable.
• 🕸 Traceable and introspectable — Every thought path is logged.
• 🔄 Designed for cognitive experimentation — Emergent behaviors, agent conflict resolution, and agent memory are on the roadmap.

Check out Getting Started for setup, installation, and your first YAML orchestration.

Want to see how it all works?
Jump into example.yaml and try running:

python test_run.py

🔜 Kafka-backed memory agent

🔜 Graph-based visualization of traces

🔜 External plugin registry for custom agents

🔜 Meta-agents (self-reflection + re-routing)

OrKa is an open research framework as much as it is a dev tool. It's built on the belief that AI cognition should be explainable, inspectable, and composable — not opaque.

This project welcomes collaboration with:

• AGI labs

• Cognitive scientists

• ML/LLM engineers

• Open-source tinkerers and thinkers

Built by Marco Somma, multipotentialite, AI engineer, and builder of cognitive tools that think out loud.