REAL COGNITION BRAIN

Building RCB
as a brain.

Arety RCB is an ambitious cognitive architecture where intelligence emerges from many interconnected thinkers—not one monolithic model.

∞cognitive nodes

N²connection challenge

01first model / 0.1

LIVE CONCEPTUAL VIEW

01 / THE IDEA

What is Arety RCB?

RCB stands for Real Cognition Brain—the underlying cognitive architecture and intelligence. It treats cognition as a distributed process where interpretation, memory, context, processing, comparison and knowledge work together as one system.

02 / FIRST IMPLEMENTATION

Arety 0.1

The first model implementing the RCB architecture. Arety 0.1 establishes the vocabulary and design principles for a full-mesh cognitive system: general nodes, dynamic decomposition, contextual memory and independent verification.

MODEL CARD

Arety RCB-1.0

Arety RCB-1.0 is the first public model built on the Real Cognition Brain architecture. It is a text model for questions, explanations, everyday conversation, calculation and translation. Each part of the brain handles a different kind of work, and they cooperate on every reply. No third-party language model writes its answers.

Model ID: arety-rcb-1.0
Developer: Arety
Release: October 2026 · research preview
Input / output: Text in, text out
Languages: Conversation in English. Word meanings and translations cover 5,000+ languages.
Knowledge: Grounded in open encyclopedic sources. People, places, organisations, works, science and everyday facts, refreshed periodically.
Memory: Remembers what you tell it, across all of your chats. Each account's memory is private to that account.
Availability: Arety Chat (web) and the Arety API (OpenAI-compatible /v1/chat/completions, personal API keys)

STRENGTHS

What it does well

Factual questions answered from its knowledge base · exact arithmetic and algebra (it calculates rather than guesses) · word meanings, synonyms and translations · remembering your name, preferences and earlier conversations · tolerating typos in names.

LEARNING

Teachable

When it does not know something it says so, and it can learn a fact you state ("the capital of X is Y"). Learned facts are kept separately from its built-in knowledge.

LIMITATIONS

Known limits

Open-ended writing (stories, long essays, code) is still basic in 1.0. Knowledge reflects its last refresh and may miss very recent events. It can misread unusual phrasings. Check important answers before relying on them.

SAFETY & PRIVACY

Your data

Signed-in chats are saved to your account and can be deleted at any time. Chats started while signed out are removed after 14 days. Nothing is shared between accounts. Passwords are stored hashed. API keys can be revoked from your account page.

Using the API and the MCP server

Programs call the REST API; AI assistants (ChatGPT, Claude, agents) call the MCP server. Full instructions with Python and JavaScript examples: tests.arety.cloud/developers. OpenAPI document: /v1/openapi.json.

# no key needed to try it (20 requests a minute per address; context kept per session)
curl https://tests.arety.cloud/v1/think -H "Content-Type: application/json" \
  -d '{"prompt":"What is 123456789 * 987654321?","session":"demo"}'

# OpenAI-compatible, with a personal key (your account's memory)
curl https://tests.arety.cloud/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" -H "Content-Type: application/json" \
  -d '{"model":"arety-rcb-1.2","messages":[{"role":"user","content":"What is the capital of Mongolia?"}]}'

# MCP server for AI assistants (streamable HTTP, no auth): tools ask_arety, reset_session, search, fetch
https://tests.arety.cloud/mcp

Create a key from Account → API keys in Arety Chat. Requests made with your key share your account's memory; keyless requests are scoped to your address and session. The API returns only the model name and the answer.

HOW IT DIFFERS

Traditional AI, LLMs and Arety RCB-1.0

There are three broad ways to build an AI system. Here is how they compare.

	Traditional AI models	Large language models (LLMs)	Arety RCB-1.0
Shape	One narrow model per task (classifier, recommender, rule system)	One very large neural network that does everything	Many cooperating cognitive parts, each good at a different kind of work
Where knowledge lives	Hand-built rules or task-specific training data	Spread across billions of weights. Hard to inspect or correct.	In an explicit knowledge base that can be inspected, updated and corrected directly
Facts	Only within its narrow task	Fluent, but can state wrong facts confidently ("hallucination")	Answers from stored knowledge, and says when it does not know
Maths	Exact, within its programmed scope	Predicts digits as text. Can be wrong on long sums.	Calculated exactly, then checked a second time
Learning new facts	Re-engineering or retraining	Costly retraining or fine-tuning	Can be taught a fact in conversation. It is used immediately.
Memory of you	Usually none	Limited to the current conversation unless extra systems are added	Built in: personal context across all of your chats
Checking itself	Fixed logic	Usually a single pass	Independent parts can reach an answer separately and are compared before replying
Hardware	Small	Large GPUs to run	Runs on ordinary CPUs. A GPU is only needed to train its text generator.
Open-ended writing	Not designed for it	Excellent	Basic in 1.0, improving with each release

In short, an LLM is an excellent writer that guesses facts from memory. RCB-1.0 looks facts up, calculates exactly, remembers you and learns from you. Its weakness is creative long-form writing, which later versions will improve.

03 / SAME INTELLIGENCE

RCB / RPTCB / RVTCB

01 / CORE

RCB

Real Cognition Brain — the underlying cognitive architecture and intelligence.

02 / PHYSICAL

RPTCB

Real Physical Technical Cognition Brain — RCB instantiated through physical computing nodes.

03 / VIRTUAL

RVTCB

Real Virtual Technical Cognition Brain — RCB instantiated through software-defined nodes.

04 / THE UNIT OF COGNITION

The cognitive node

Each node is a participant in cognition. Categories describe current strengths, not permanent jobs.

LanguageMemoryContextProcessingComparisonInformationWeb / Knowledge

05 / FUNDAMENTAL CONNECTIVITY

The full-mesh brain

Full mesh is fundamental. Every node remains capable of communicating with every other node. Participation in a particular thought remains dynamic.

06 / DYNAMIC COGNITION

How a thought happens

A task can be decomposed dynamically across language, processing, context, comparison and knowledge nodes. The architecture does not require one permanent route for every kind of thought.

07 / MEMORY ARCHITECTURE

Memory & context

USER CONTEXT

Persistent context

Long-term user information and memory.

CHAT CONTEXT

Conversation context

Messages and working context loaded for the current conversation.

ENGINEERING

Memory tiers

Local working state, shared task context and persistent knowledge are proposed implementation layers.

08 / INDEPENDENT CHECKS

Comparison & verification

Nodes can produce different answers or reasoning paths. Comparison evaluates semantic equivalence, logical consistency, factual grounding and constraints rather than reducing cognition to simple majority voting.

09 / PARTICIPATION

Semantic routing

Meaning, complexity, capability and runtime state may determine which nodes actively participate in a thought, while full-mesh connectivity remains available.

10 / INFORMATION INFRASTRUCTURE

Information & web knowledge

Future knowledge systems must address storage scale, indexing, freshness, timestamps and hashes, source metadata, licensing and the relationship between information and its context.

11 / IMPLEMENTATION LAYERS

Python + C++ + JSON

PYTHON

Cognition & reasoning

Interpretation, task division, node logic and higher-level processing.

C++

Performance infrastructure

Data, memory, communication, networking and low-level operations.

JSON

Logical protocol

Node-to-node communication and interchange representation.

12 / ENGINEERING FRONTIER

The networking challenge

Full-mesh communication creates N(N−1)/2 pairwise relationships. Possible future optimization areas include shared memory, IPC, high-speed Ethernet, RDMA, zero-copy, compact serialization, caching, batching and asynchronous communication.

13 / HONEST STATUS

Implemented vs proposed

CORE ARCHITECTURE

RCB identity, full-mesh principle, general cognitive nodes, dynamic decomposition, context distinction, comparison as cognition, substrate independence and Python/C++/JSON roles.

RESEARCH / PROPOSED

Advanced routing policies, large-scale web indexing, RDMA, zero-copy, batching, hybrid deployments and scoped consensus tools.

14 / QUICK REFERENCE

FAQ / glossary

Is RCB a conventional LLM?

No. RCB describes a distributed cognitive architecture rather than a single monolithic model.

Does full mesh mean every node participates in every task?

No. Participation is dynamic; full mesh means nodes remain capable of communicating with one another.

Is Arety claiming AGI or consciousness?

No. Arety is an architecture under development and makes no such claim.

What is the difference between Chat Context and User Context?

Chat Context is current-conversation working context; User Context is persistent information and long-term memory.

Building RCBas a brain.