Architecture overview

The intelligence layer.
How it works.

ClusterOS is the coordination infrastructure for regional economies — the layer that makes actor behaviour, diagnostic intelligence, and steward decision-making work together at scale. Its defining purpose is a single one: making complexity cheaper to operate.

This page explains how the architecture delivers that — how the layers are structured, what each design choice means, and why the architecture itself is the product.


01

System Architecture

The diagram below shows how ClusterOS is structured. Read it from the bottom up: external data and the sovereign database form the foundation; the intelligence layer sits above them; what actors experience is generated at the top. The connections between layers are the product.

What is a cluster?

Clusters are not imposed categories. They emerge from the diagnostic's evidence analysis and are validated with the steward. A cluster is a thematic innovation domain defined by the actors and dynamics actually present — not by a taxonomy someone decided in advance. Clusters overlap by design. A single actor can belong to multiple clusters. A single piece of evidence can inform multiple cluster diagnostics. The platform reflects real economic structure rather than enforcing artificial silos.

What is MCP?

Model Context Protocol is an open standard that allows an AI model to call structured tools — read data, query state, trigger actions — rather than relying on pre-trained knowledge. The backend exposes the ecosystem database as a defined set of named tools. The AI calls those tools at runtime to assemble what it needs for a specific actor at a specific moment. Intelligence is not hardcoded into the interface. It is reasoned fresh from live data on every call. When the AI model improves, everything the platform produces improves automatically.

Where do signals come from?

Signal ingestion is automated, not manual. Autonomous bots — scheduled AI agents — monitor external sources continuously: government support directories, community service APIs, publication databases, professional identity systems, grant announcements. They write structured signals into the sovereign database on a defined cadence. Actors generate additional signals through their own platform behaviour: posting, connecting, progressing journey steps, responding to RFIs. The system arrives pre-populated. No actor is ever asked to fill a blank page.

Why do actors contribute?

They do not need to think of it as contributing. Every actor uses the platform in their own self-interest — founders find customers and funding, anchors find capability, researchers find commercial relevance, stewards understand what is actually happening in their ecosystem. Every selfish action generates a typed signal that makes the system more intelligent for everyone else. The incentive is immediate and personal. The benefit to the ecosystem is structural and automatic. No actor needs to care about the ecosystem for the ecosystem to function as a system.

ACTOR EXPERIENCE INTELLIGENCE PROTOCOL DATA LAYER EXTERNAL SIGNALS GOVERNANCE FOUNDER JOURNEY Personalised pathway Step unlock logic Matched connections Support programme alerts Generated at runtime ANCHOR JOURNEY Capability scouting RFI signal management Supply chain visibility No premature disclosure Generated at runtime RESEARCHER JOURNEY Demand signal translation Commercial opportunity map IP boundary controls Research partner matching Generated at runtime STEWARD INTERROGATION System health signals (4 per cluster) Stall detection · Stack analysis Open-ended diagnostic interrogation Leverage hypothesis surfacing Not a dashboard — questions nobody designed for AI REASONING ENGINE Journey Construction Signal Propagation Role-Aware Reframing Actor Matching Stall Detection Steward Interrogation The AI is the runtime, not a feature. Every actor experience is constructed here, from live data, on demand. MODEL CONTEXT PROTOCOL (MCP) — INTELLIGENCE SUBSTRATE get_actor_profile() get_cluster_signals() detect_stalls() get_matched_connections() propagate_signal() get_steward_diagnostic() get_journey_state() search_support_programmes() get_open_funding() ingest_evidence() enrich_actor_profile() get_leverage_hypotheses() Each tool is a structured call from the AI to the live database. Intelligence is not pre-baked — it is assembled from real data, at the moment it is needed. SOVEREIGN ECOSYSTEM DATABASE Supercluster · Clusters · Groups · Actors Actor profiles · Signal history · Journey state · Evidence base Diagnostic outputs · Stall records · Leverage hypotheses One dedicated instance per regional EDA. Data does not leave the regional tenant. No cross-region data sharing. Full role-based access control. Sovereign by design. AUTONOMOUS BOT NETWORK Evidence Ingestion Bot Signal Monitoring Bot Actor Enrichment Bot Health Monitoring Bot Scheduled AI agents. Write structured signals into the database on a defined cadence. Cold-start solved: the platform knows the ecosystem before any actor joins. GOVT SUPPORT DIRECTORY Live programme listings Funding routes by stage Advisor networks Ingested weekly via scrape · Open licence COMMUNITY SERVICES API Wellbeing services by location Founder peer networks Community assets REST API · CC BY 4.0 PUBLICATION DATABASES Research outputs Grant announcements Patent filings Feeds researcher journey signals PROFESSIONAL IDENTITY Professional profile import Research identity (ORCID) Behavioural enrichment Progressive — never compulsory FEDERATED GOVERNANCE LAYER Supercluster › Cluster › Group › Actor Identity-first · Role-based access · Full audit trail Institutional autonomy preserved at every level Data residency is jurisdictional. Each EDA operates a sovereign instance. The platform does not centralise regional data. Active signal flow Data retrieval (MCP call) Governance / auth boundary
The load-bearing structure

The Sovereign Ecosystem Database.
Everything else depends on it.

Every actor journey, every diagnostic output, every signal, every stall record — all of it lives in a single, dedicated database per regional EDA. This is not a shared multi-tenant cloud service where your data sits alongside everyone else's. Each regional deployment is a sovereign instance. Data does not leave the regional tenant. Intelligence is generated locally from local data.

The hierarchical structure of that database is what makes the rest of the platform possible. Superclusters contain clusters. Clusters contain groups. Groups contain actors. Actors generate content and signals. Every relationship, every interaction, every piece of evidence is stored with full provenance — who created it, when, in what context, with what permissions.

This structure is why the platform can route a procurement RFI from an anchorA large established actor — corporate, university, hospital, or public body — whose scale shapes the conditions other actors navigate. Some anchors also carry stewardship responsibility for the cluster. to three matched founders without broadcasting it to four hundred cluster members. The database knows the relationships. The AI reasons against them. The result is intelligence, not noise.

Supercluster Cluster Group Actor Signal + Journey state + Evidence base

02

One event. Four reframings.

A single actor does something selfish. In a broadcast system that signal goes to everyone and is ignored by most. In the ClusterOS infrastructure, the same signal is typed, matched, and reframed for each actor based on what it actually means to them. Same underlying data. Four different intelligence surfaces.

The triggering event
Anchor · Rolls-Royce Scotland
Posts an open RFI: seeking advanced composite materials capability for next-generation turbine component. 90-day procurement window.
Founder · composite materials startup
"You match this RFI. The application window closes in 47 days. Two preparation steps are available in your journey before you're ready to respond."
Acts on a specific, timely signal — not a general announcement buried in a newsletter.
Researcher · materials science, University of Glasgow
"This procurement need aligns with your published work on fibre-reinforced polymers. Two founders in your cluster are potential commercial routes. A collaboration pathway is available."
Published research reaches live commercial demand without the researcher monitoring procurement feeds.
Steward · Scottish Advanced Manufacturing cluster
"Buy-side demand signal in Advanced Manufacturing. Three founders have matching capability. Bridging score: 34/100. Possible Mediating stall — direct connections are not forming. Intervention available."
Sees a systems-level opportunity with a named diagnostic pattern — not just an activity log.
Second anchor · BAE Systems
"A founder in your adjacent capability area has reached validation stage. Their composite materials profile is now visible to matched anchors. No commercial disclosure required at this stage."
Learns about relevant capability without premature commitment — timing managed by the infrastructure.
03

Why the architecture is the product

The architecture is built around six deliberate design choices. Each one addresses a specific failure mode in how regional ecosystems are typically managed — and together they define what makes ClusterOS infrastructure rather than software.

P.01

Intelligence lives in the protocol, not the interface

Most platforms embed decisions in their UI — what metrics to show, what to flag, what comparisons to draw. Those decisions are made once and frozen in code. ClusterOS inverts this: the interface is a thin, stable shell. All the thinking happens in the protocol layer, called fresh from live data for each actor at each moment. When the underlying AI model improves, every journey and interrogation improves automatically — without rebuilding the interface.

P.02

Every actor's self-interest feeds the whole system

Actors do not need to care about the ecosystem. They care about finding customers, funding, research partners, and talent. Every action they take in their own interest — posting, connecting, progressing a step — generates a typed signal that propagates to other actors. The platform makes self-interest productive for the system. This is the CAS principle externalised into infrastructure: coordination without requiring anyone to coordinate.

P.03

Sovereign data by architecture, not policy

Each EDA operates a dedicated, isolated instance. Data does not leave the regional tenant. The AI reasons from patterns within the instance — it does not train across regions or expose one region's data to another. Data sovereignty is not a compliance layer added on top. It is built into the data architecture. The EDA owns its collective intelligence permanently.

P.04

External data solves cold start without burdening actors

New platforms fail because they ask actors to populate them before there is anything to see. ClusterOS solves this at the architecture level: autonomous bots ingest external data — support programmes, service directories, publications, announcements — before any actor joins. The first founder to onboard encounters a platform that already understands their ecosystem context. They are never looking at a blank page.

P.05

Steward interrogation, not steward dashboards

A dashboard answers questions someone anticipated. ClusterOS also surfaces the metrics that matter — throughput, conversion, anchor engagement, stall configuration — but it goes further. A steward can interrogate the intelligence layer directly: why is throughput stalling in this cluster specifically, which founders most resemble our successful exits at the same stage, what would one new actor introduction unlock. The architecture makes both structured metrics and open questions answerable from the same data.

P.06

The diagnostic runs continuously, not as a report

Most ecosystem diagnostics are documents — produced once, read once, shelved. ClusterOS runs its 5-stage diagnostic pipeline (Evidence → Patterns → Stalls → Stacks → Leverage) continuously as the bot network ingests new signals and actor behaviour generates new data. The platform gets smarter as the ecosystem becomes more active. Complexity becomes manageable rather than something to be periodically summarised and filed.

"ClusterOS reframes regional development from building more initiatives to making complexity cheaper to operate. That is its defining purpose — and its enduring value."
ClusterOS™
A Cognitive Operating System
for Regional Economies

Digital Architecture

See how this maps to your ecosystem

Identify your stalls, detect the reinforcing dynamics, and see which layers of ClusterOS address your specific configuration.

Start your diagnostic →