Independent Research & Technology Lab

Autonomous Intelligence.
Frontier Research.
Resilient Systems.

Duara Cortex is a United Kingdom research organisation studying the architectural foundations of frontier systems. We help businesses and individuals put intelligent systems to work. You bring us a problem in your business or your project, we study it, and we give you a written plan you can act on: what to build, how it should fit together, and what introducing it will actually take. What you do with it is yours. Build it in-house, hand it to a contractor, or bring it back to us.

Explore Research Areas Read the Charter

// ENGAGEMENT ACTIVE LAB

You Bring

A problem in your business or project

We Do

Research, analysis, and system design

You Receive

A written plan: what to build, how it fits together, what it takes

Ownership

Yours. Build it anywhere.

Structure

Non-profit, England and Wales

Our open source code

github.com/Duara-Cortex ↗

[02 // APPROACH]

How We Work

People come to us knowing that intelligent software could help them, but not what to build, what it would cost, or whether it would work at all. The technology moves quickly, most of it is written up for people who already understand it, and the cost of guessing wrong is high. That gap is the reason this organisation exists.

So we treat every engagement as a piece of research rather than a sales conversation. We study the problem properly, draw on a network of researchers and engineers assembled to suit the work, and hand back a plan you own outright. Build it in-house, give it to a contractor, or bring it back to us. The recommendation is the same either way.

Our Working Principles

01 PRINCIPLE

We Start With Your Problem

Not with a technology. Many problems that look like they need a smart system turn out to need something simpler, and we will tell you when that is the case. The study comes before the recommendation.

02 PRINCIPLE

The Plan Is the Deliverable

You receive a written document you can act on without us: what to build, how the parts fit together, what it will take to introduce, and what could go wrong. It is written to be handed to whoever ends up building it.

03 PRINCIPLE

Nothing Is Finished Until It Runs

We do not consider an idea proven until it has been built as working software that someone outside the organisation could pick up, depend on, and understand. Our own tools are published openly for exactly that reason.

04 PRINCIPLE

What Is Useful Gets Published

When a problem turns out to matter beyond the person who brought it to us, we write it up and publish it, including the attempts that did not work. Your commercial details stay yours. What is general becomes available to everyone.

[03 // INQUIRIES]

What We Study

Three problems come up in almost every engagement, so they are where our research concentrates. They are connected: a system that reasons well but cannot be run reliably is no use to anyone, and a reliable system with nothing to reason about is not worth building. The work below is where those two ends meet.

INQ-01 RESEARCH PROGRAMME

Systems That Run Without Supervision

Multi-Agent Consensus Memory Synthesis Reflective Planning Distributed Coordination

Software that acts on its own has to decide what to do next, remember what it has already done, and settle disagreements when several parts of it want different things. Get this wrong and the system needs a person watching it constantly, which defeats the point. We study how autonomous systems coordinate without human arbitration, how they decide what to keep and what to forget over long periods, and how they check their own reasoning before committing to a course of action.

INQ-02 RESEARCH PROGRAMME

Systems That Do Not Fall Over

Go Python TypeScript Concurrent State Sandboxing Low-Latency Runtime

A system is only as trustworthy as the infrastructure running it, and intelligent software fails in ways ordinary software does not. We work on the engineering underneath: keeping state consistent when several processes act at once, containing what a system is allowed to do so mistakes stay contained, and keeping running costs low enough that the thing is still worth operating a year later. Principally in Go, Python, and TypeScript.

INQ-03 RESEARCH PROGRAMME

Making Sense of Information You Already Have

Relational Models Formal Verification High-Dimensional Search Retrieval-Augmented Reasoning

Most organisations hold far more information than they can use. Documents, records, and notes accumulate in places nobody searches, and a system given raw material in that state produces confident answers nobody can check. We study how to structure information so it can be queried, verified, and traced back to where it came from, and how to find the relevant part of a large body of knowledge at the moment it is needed.

[04 // REPOSITORIES]

Applied Software & Projects

Duara Cortex treats its published software as the primary evidence of its research, not as a secondary output. All formal releases are maintained under the organisation's GitHub account: github.com/Duara-Cortex

The organisation is at an early stage: formal releases under the Duara Cortex account are forthcoming, as existing exploratory work is formalised, documented, and migrated into the organisation's public repositories. That existing body of work already gives a reasonably direct sense of the research programme, and includes:

REPOSITORY MIGRATION

Exploratory codebases are actively being consolidated and prepared for public Apache-2.0 release under @Duara-Cortex.

Exploratory Implementation

Self-Hosted Knowledge Vault

A system for aggregating and structuring personal and organisational knowledge — a practical testbed for the semantic-synthesis research described above.

Markdown Knowledge Graph Local-First

Exploratory Implementation

Orchestration Daemon

High-precision daemon for scheduling and executing agentic routines within precise time windows — early infrastructure for the coordination problems in agentic workflows.

Go Concurrency Agent Scheduling

Algorithmic Prototype

Deterministic Hashing Construction

Algorithmic construction built on recursive geometric expansion — an example of deterministic-systems research applied to a concrete cryptographic problem.

Cryptography Deterministic Algorithms Geometry

Research Testbed

Long-Term Memory Layer

A graph representation mapping notes and their links weighted by usage and connection density — a direct precursor to our published work on memory synthesis.

Graph Theory Memory Synthesis Vector Semantics

Active Tooling

Remote Orchestration API

An API for uploading, managing, and triggering agentic skills and tools from external harnesses.

TypeScript APIs Skill Harness

PUBLIC REGISTRY

Duara-Cortex on GitHub

Follow our public repositories, issues, open RFCs, and benchmark pipelines as implementations are formalised.

Explore repositories →

[05 // STRUCTURE]

Why We Are a Non-Profit

Most people who advise you on technology have something to sell you afterwards. The plan they hand over tends to describe the thing they were always going to build. We wanted to remove that incentive entirely, so Duara Cortex is registered in England and Wales as a Company Limited by Guarantee. It has no shareholders and no share capital, which means there is nobody waiting on a return and no reason for our advice to point anywhere but at the answer.

Written into our governing documents is an asset lock. Anything the organisation earns or owns can only be spent on the work itself, and cannot be paid out to any private individual, including if the company closes. Fees from advisory work and any grants we receive go back into research, tooling, and the software we publish openly. This is a legal constraint on the company rather than a promise on a website, and it is what lets us give the useful parts of our work away.

Advice With No Agenda

We have nothing to sell you at the end of it. The plan recommends what we think you should do, not what we would rather build.

No Owners

No shareholders and no share capital. Work is paced by the problem in front of us rather than by a quarterly target.

Everything Reinvested

Every fee and grant goes back into research, tooling, and open software. Nothing is paid out to private interests, by law.

[06 // DISPATCH]

Inquiries & Collaboration

Duara Cortex welcomes contact from academic researchers working in adjacent areas, from open-source contributors interested in the software described above, and from organisations considering an advisory or consulting engagement. Correspondence is read directly by the research team rather than routed through a general inbox.

Academic & Technical

Joint empirical investigations, paper reproductions, or open architectural critique.

Advisory & Systems Review

Targeted consulting engagements for agentic runtime safety and semantic knowledge architectures.

Direct Channels RECEPTIVE

Electronic Mail

contact@duaracortex.com

Correspondence & Registered Office

Duara Cortex Limited
66 Paul Street
London, England, EC2A 4NA
United Kingdom

Codebase & Research Issues