Radiolaria OS: Computing, Assembled Around Intent

Abstract

Radiolaria OS is an application execution kernel for organizing practical work around human intent. It brings together language models, programs, services and numerical tools while separating semantic proposals, bounded execution, evidence and decisions by local resource owners. Its purpose is to support applications that can respond to changing requirements and resources without giving an individual model unrestricted control over the activity.

The reference materials illustrate complementary aspects of this architecture. Football demonstrates exchange of computed material between independent owners and a separately authorized mock booking. Wedding connects model-returned changes to computation and follows recorded quantum results through validation and consumption. An ephemeral workspace demonstrates selective continuation after a resource disappears. Additional materials describe time, memory, provenance, experience-based allocation and independently examined model-authored extensions.

This record archives the eight assets of the GitHub release radiolaria-os-reference-g6b6f-v01, first published on 29 September 2026. It includes a whitepaper, an editable presentation and its PDF, a Geometry Atlas, an explanatory XML Reader, a historical reader-evaluation bundle, and the complete declared public source export in XML and ZIP formats. The explanatory Reader is intended for architectural understanding; the large source export is intended for indexed or tool-assisted inspection, not a single model prompt.

Akhtanin Andrii led the architecture and research. AI tools, including ChatGPT/Codex and Gemini, assisted implementation, analysis, documentation, translation and visual preparation. Recorded model interactions are identified as experimental material, not human peer review. The human author remains responsible for the work. The documentation distinguishes controlled examples, historical provider responses, mock effects, retained evidence and untested deployment assumptions.

Licensing is component-specific. Code is licensed under AGPL-3.0-only. New explanatory author text and original diagrams are licensed under CC BY 4.0. Verbatim code excerpts, historical material, third-party content and embedded font subsets retain their applicable notices. These are not alternative blanket licenses for the entire deposit. See RIGHTS.md and the license notices in the source bundle.