GNUS.ai and GCS: Architecture at a Glance

Reviewed: October 9, 2026. Conceptual architecture, not a deployment diagram or proof of completed integrations. This is the first-read technical guide for developers, customers, and reviewers.

The idea in two paragraphs

GNUS.ai combines SuperGenius, a distributed processing, trust, and GNUS settlement platform, with Genius Cognitive System (GCS), a modular approach to AI work. GCS is intended to select a Semantic Core, compact Expert Language Models (ELMs) and other specialist capabilities, authorized memory, tools, and checks that each task needs. The goal is to avoid paying for a large generic model on every step when a smaller qualified specialist could suffice.

GCS plans the work. SuperGenius owns native processor selection, SubTask scheduling, peer networking, and accounting when a request actually needs public distributed execution. A local or tenant-private request might never use a public node, escrow transaction, or on-chain proof. A conceptual arrow does not imply that each component already runs in one complete production stack.

flowchart TD
  A["Application / API / SDK"] --> B["GCS request policy + planner"]
  B --> C["Semantic Core + selected ELM/EJM specialists"]
  B --> D["GAML / retrieval / context (governed memory)"]
  C --> E["Cognitive checking + synthesis"]
  D --> E
  B --> F{"Execution mode?"}
  F --> G["Local or private compute"]
  F --> H["Planned GCS-to-SuperGenius ELM bridge"]
  H --> I["ONE funded native SuperGenius ELM job"]
  I --> J["SuperGenius SubTasks + worker selection"]
  J --> K["Processing + result and settlement"]
  G --> E
  K --> E
  E --> L["Response / authorized memory update"]

Boundaries readers should not confuse

Question	Responsible system	Caution
What experts, memory, and checks are needed?	GCS orchestration	The broad roadmap is not equivalent to a deployed GCS controller.
Which native worker gets a distributed task?	SuperGenius processing	Do not implement a second SuperGenius scheduler per ELM.
Did a worker run the declared model and kernel?	Planned GCS Execution Integrity System (EIS), with runtime evidence	EIS specification is still draft and not a published cross-GPU attestation result.
Is an answer correct?	GCS grounding, semantic verification, arbitration	Not the same as matching execution hashes or ledger consensus.
Is network state authorized, and who gets paid?	SuperGenius consensus and settlement	Neither implies answer quality.
Can a consumer access /v1 Chat Completions?	Planned OpenAI-compatible GCS gateway	v1.1 Phase 6 is not a publicly operational service.

Real implementation versus the plan

Built software exists, including SuperGenius processing/ledger sources, NEO-SWARM C++ inference/routing modules and tests, and GCS Chat C++/Flutter messaging foundations. The full cognitive system, governed GAML memory, EIS and OpenAI-compatible retail API are not established as deployed and integrated in the current system workstream. Consult Implementation Status, Performance and Benchmarks, and Document Index for evidence and limitations.

Canonical context: GCS System Overview · Native bridge specification · GNUS Platform Status · SuperGenius issue #369.