Architecture Overview
See how the cognitive network fits into Squad’s platform architecture.
What is a Cognitive Network?
Most enterprise AI falls into one of two camps. RPA automates clicks — it follows scripts, breaks when the UI changes, and understands nothing about the work it’s doing. AI agents are smarter, but they operate in isolation: one model, one prompt, one task, with no shared understanding of your organisation.
Both approaches share the same limitation: intelligence is localised. There’s a central brain — the LLM, the analytics team, the single AI tool — and everything else is just dumb plumbing that feeds it. It’s like having a brain but no nervous system: you can think, but you can’t feel your hand on the stove.
A Cognitive Network changes this fundamentally. It is your organisation’s nervous system.
From central brain to distributed cognition
Most AI architectures follow the same pattern: raw data flows through pipes into one brain, which produces an answer. All the intelligence lives in one place. Everything else — your data integrations, your document stores, your process maps — is inert infrastructure. It connects and moves information, but it doesn’t understand any of it.
In the human body, cognition isn’t centralised like this. Your spinal cord handles reflexes without involving the brain. Your gut has its own neural network. Your immune system “remembers” threats independently. The brain coordinates, but intelligence is distributed throughout the system. That’s what makes it fast, resilient, and able to handle complexity at scale.
A cognitive network applies this same principle to your organisation. Because a large portion of the inference is stored and crystallised directly in the graph — entity classifications, resolved relationships, proven patterns, domain ontologies — “thinking” exists everywhere in the network, not just at the centre. It’s as if the nerves themselves had neurons.
When a new document is ingested, it doesn’t need to phone home to a central brain to understand what it’s looking at. The local graph structure already carries enough crystallised understanding to classify, resolve, and contextualise. The central reasoning engine orchestrates, but every node in the network already knows things.
The upgrade to the semantic layer
A semantic layer maps and connects your data — it’s the wiring diagram. A cognitive network is what happens when you run intelligence through the wiring.
Where a semantic layer gives you a structured map, a cognitive network applies cognition at every point in that map:
- During ingestion: entities aren’t just extracted — they’re classified, resolved against known entities, and placed into an evolving domain ontology that adapts to your organisation without manual schema work.
- During retrieval: queries aren’t just matched — they’re interpreted through your organisation’s context, domain vocabulary, and proven patterns.
- During execution: workflows aren’t just automated — they follow your policies, procedures, and guidelines, with built-in uncertainty detection and human-in-the-loop controls.
- During learning: every interaction strengthens the network. Approved patterns become reusable templates. Corrected outputs improve future accuracy. The system gets smarter with use.
The critical differentiator is crystallisation. Every time Squad resolves an entity, classifies a type, or approves a pattern, that intelligence is deposited into the graph — not kept in a model’s weights or a prompt. It becomes structural knowledge that any future process can use without re-inferring it. The graph gets smarter, not just the model.
Why this matters for high-stakes work
In defence, financial services, and regulated enterprise, “close enough” isn’t good enough. These environments demand:
- Measurable accuracy — not best-effort generation, but outputs you can quantify and audit
- Full traceability — every decision traceable back to source data, through every reasoning step
- Policy adherence — AI that follows your rules, not just its training data
- Progressive trust — a system that earns confidence through demonstrated performance, not promises
A chatbot can’t deliver this. An isolated agent can’t deliver this. It requires a nervous system — one where intelligence is embedded throughout the organisation, not bolted on at one point.
How Squad builds a Cognitive Network
Squad’s cognitive network is built on three reinforcing systems:
1. Structured knowledge
Every piece of data ingested into Squad passes through the Universal Semantic Encoding Pipeline (USEP) — a staged extraction cascade that classifies entities using the POLE+O framework (Person, Organisation, Location, Event, Object) extended with cognitive primitives (Concept, Tool, Procedure, Fact).
Entity types aren’t rigid. They crystallise organically as data is ingested: clusters form in the embedding space, stable patterns are detected, and new classifications emerge automatically. A defence logistics deployment develops different entity types than a financial services one — without manual configuration.
| Type | Description | Examples |
|---|---|---|
| Person | Individual people | Employees, contacts, public figures |
| Organisation | Companies, institutions, groups | Suppliers, partners, regulatory bodies |
| Location | Places and geographic features | Facilities, regions, addresses |
| Event | Occurrences and incidents | Meetings, incidents, deliveries |
| Object | Products, artifacts, works | Equipment, documents, assets |
2. Entity resolution across sources
When the same entity appears across multiple documents and data sources, Squad resolves them to a single canonical representation through progressive consolidation: deterministic matching first (handling ~75% of cases), followed by semantic resolution for ambiguous cases. All merges are soft and reversible — medium-confidence cases are flagged for human review rather than auto-merged.
This is what makes it a network, not just a database. Cross-source resolution means your knowledge graph reflects reality: one supplier, one policy, one asset — regardless of how many systems reference it by different names.
3. Cognition at every layer
The network is powered by Squad’s Active Inference Model (AIM) — a reasoning engine built on the Free Energy Principle, Bayesian inference, and dual-process cognition. AIM reads from and writes to the knowledge graph at every stage of processing, meaning:
- Retrieval is informed by the full context of your organisation’s knowledge
- Reasoning accounts for uncertainty and confidence levels
- Actions follow your documented policies, processes, and procedures
- Outputs are auditable end-to-end, with full provenance
Cognitive Networks vs. the alternatives
Most AI companies are selling a bigger brain. Squad is selling a smarter nervous system.
| RPA | AI Agents | Cognitive Network | |
|---|---|---|---|
| Where intelligence lives | Nowhere — follows scripts | In the model | Distributed across the graph |
| Understands context | No | Partially — single-task context | Yes — full organisational context |
| Cross-source intelligence | No | Limited | Yes — unified entity resolution |
| Learns from use | No | Sometimes | Yes — every interaction strengthens the network |
| Policy adherence | Hardcoded rules | Prompt-based | Structural guardrails with human-in-the-loop |
| Handles ambiguity | Breaks | Hallucinates | Detects uncertainty, escalates to humans |
| Audit trail | Action logs | Varies | Full provenance from source to output |
What this enables
The cognitive network architecture enables capabilities that aren’t possible with disconnected tools:
Entity-centric investigation: start from any entity and explore its connections across every data source — which documents mention it, what other entities it co-occurs with, how it fits into broader community structures.
Supply chain and network mapping: trace relationships between organisations, locations, and assets. Identify single points of failure and understand the structure of complex operational networks.
Digital twin construction: model real-world assets, systems, and processes as a connected graph with typed relationships — a queryable digital representation that stays current as new data is ingested.
Cross-source intelligence: resolve the same real-world entity across different data sources and document types. Build a unified view even when different sources use different names, abbreviations, or identifiers.
Data lineage and provenance: trace how information flows across your organisation. Every entity carries full source provenance, enabling you to answer “where did this fact come from?” across the entire knowledge base.
AIM: Active Inference Model
The reasoning engine that powers cognition across the network.
Memory Architecture
How Squad organises knowledge across orthogonal memory systems.
Data Ingestion
How data enters and is processed by the cognitive network.