Marco, an ex-ethologist turned AI systems engineer, created OrKa (Orchestrator Kit for Agents) to address frustrations with existing black-box AI tools like LangChain and AutoGPT. He sought an open-source framework that allows users to wire cognition like circuits, focusing on traceability, memory, and reproducibility. Inspired by cognitive processes, OrKa enables users to create YAML-defined cognition graphs with features like fork/join execution, router agents for confidence-weighted branching, and scoped memory. The system supports local and remote LLMs and includes a React-based visual UI. Benchmarks indicate efficiency, with a simulated average latency of ~7.6 seconds per agent and stable performance metrics. Marco invites feedback on OrKa’s design and future direction, questioning whether it should focus on infrastructure or evolve into a hosted service. Links to explore OrKa further are provided, including its documentation, GitHub repository, and UI on Docker.
Introducing orKa-reasoning: A Modular Framework for Explainable AI Agent Orchestration
