For Alok Aggarwal & the Scry AI engineering team  ·  Concept brief

The runtime for Collatio, Auriga, and Concentio at enterprise scale.

Scry AI runs three named platforms — Collatio® (IDP), Auriga® (Conversational AI), Concentio® (IoT) — backed by 60+ proprietary CognitiveBricks for Fortune 500 customers including Microsoft, NEOM, Cisco, Wells Fargo, Kaiser Permanente, and Wolters Kluwer. That shape — per-customer isolated runtimes, multi-modal AI pipelines, on-prem-friendly compliance, global enterprise deployments — maps directly onto Cloudflare's developer platform. You're already on Cloudflare at the edge. This is what the platform layer looks like.

3
Named AI platforms
60+
CognitiveBricks algorithms
2014
"Before AI was mainstream"
330+
Cloudflare POPs
The thesis

Scry AI's edge isn't AI hype.
It's the discipline to ship three platforms, 60+ algorithms, and zero hallucinations.

Most "AI companies" run a single LLM wrapper. Scry AI runs three independent platforms with shared infrastructure — a library of proprietary CognitiveBricks, an IQ-SMART covenant model, and on-premises-capable deployments for regulated enterprises. That discipline maps directly onto Cloudflare's primitives. Not "kind of" — exactly.

What we noticed in your stack

scryai.com is already on Cloudflareserver: cloudflare, cf-ray, cf-cache-status: HIT in every response. That's the easy part. The harder observation: your x-gateway-cache-key and x-gateway-cache-status headers suggest you've built (or licensed) your own caching gateway on top — which means you understand edge value. Mail on Microsoft 365. MongoDB confirmed. Apps deployed per-customer. The conversation isn't "should you use Cloudflare." It's "Cloudflare's developer platform is the natural home for Collatio + Auriga + Concentio as you scale to the next 50 enterprise logos."

Your three platforms

Each maps to a different center of gravity in Cloudflare's developer platform.

Reading your /collatio, /auriga, and /concentio product pages, the dominant Cloudflare primitive shifts per platform — but Workers for Platforms is the connective tissue.

Platform 01

Collatio®

Intelligent Document Processing

Document ingest at any volume, any format. Reconciliation. KYC/KYB. Loan ops. Investment statements. Contract intelligence. Each enterprise customer gets isolated runtime + their own document corpus.

Workers for Platforms R2 Workers AI Vectorize
Platform 02

Auriga®

Conversational AI on your data

Multi-modal queries (chat + voice + avatar) over enterprise data with source-linked traceability. Knowledge Agent. Customer Support 360. Analytica. CreditIQ. Multi-LLM by definition (your Realtime Intelligence ties to Parsons Corp data).

AI Gateway Workers AI Vectorize Durable Objects
Platform 03

Concentio®

AI-first IoT Platform

Real-time edge computing. Digital twin modeling. Multi-protocol device interoperability. City Intelligence, Smart Utilities, Connected Worker, SceneTrack CCTV, Drone-based Infra. This is textbook edge-native workload.

Workers Durable Objects Workers AI Workflows
IQ-SMART Covenants

Your 7 principles, mapped to Cloudflare primitives.

Each letter of IQ-SMART has a direct Cloudflare counterpart. Read across.

I
Integrated
Workers integrate behind any firewall via Tunnels + Connectors
Q
Quality
AI Gateway audit logs + traceable retrievals via Vectorize
S
Secure
Workers for Platforms = per-customer isolated runtimes
M
Monetary
R2 zero-egress + per-request pricing eliminates AWS surprises
A
AI-Powered
Workers AI catalog + AI Gateway multi-provider routing
R
Ready
wrangler deploy ships the whole stack in one step
T
Timeliness
330+ POPs = sub-50ms responses globally, no regional hop
Value plays

Eight things Cloudflare changes for Scry AI.

Ranked by impact-per-effort for your specific workload shape — multi-platform AI for Fortune 500 enterprises with on-prem requirements.

01 — Flagship

Per-customer isolated runtimes with Workers for Platforms

Microsoft, NEOM, Cisco, Wells Fargo, Kaiser Permanente — each runs their own configuration of Collatio + Auriga + Concentio. Workers for Platforms dispatch namespaces give you one isolated worker per enterprise. Each customer's CognitiveBricks combination, RAG corpus, LLM policies — fully isolated, individually metered. The "Secure" in IQ-SMART, in primitive form.

Workers for Platforms Dispatch Namespaces
Direct map: per-customer enterprise isolation
02 — Auriga

AI Gateway for multi-LLM, audited inference

Auriga claims "verifiable answers with source traceability for compliance and trust" — that's an AI Gateway promise. Sit it in front of whatever LLM mix you run (Azure OpenAI for Microsoft customers, self-hosted for on-prem, Claude/GPT where allowed). Per-customer cost attribution. Semantic cache on repeated enterprise queries. Full audit log per call. The "Quality: zero hallucinations" claim becomes evidentially demonstrable.

AI Gateway Semantic Cache Audit Logs
See calculator below ↓
03 — Collatio

R2 for the enterprise document corpus

Collatio handles every doc format an enterprise throws at it — PDFs, scans, tables, charts, schematics. That's a lot of binary storage. R2 with zero egress means customers querying their archive, auditors running compliance reviews, and your own re-training passes don't get hit with egress fees. S3 egress is the silent margin tax on doc-processing platforms; R2 eliminates it.

R2 Zero Egress S3-compatible API
Typical 40-60% storage TCO reduction
04 — Concentio

Durable Objects per IoT asset / scene

Concentio's "real-time edge computing and digital twin modeling" is the textbook Durable Objects workload. One DO per asset (drone, CCTV camera, smart meter, connected worker, vehicle) — strong consistency, geo-routed to the nearest POP, hibernate when idle, resume on event. Replaces the SCADA-style stateful service tier with a managed primitive.

Durable Objects Storage API Workers AI
Native digital-twin runtime
05 — Auriga RAG

Vectorize for the Enterprise Knowledge Agent

Auriga's Enterprise Knowledge Agent is RAG over the customer's data. Vectorize gives you a managed vector DB at edge latency, per-customer-tenant isolated, with sub-30ms semantic queries. Pair with Workers AI embedding models for the indexing pipeline. No external vector DB to operate, no Pinecone or Weaviate bill.

Vectorize Workers AI Embeddings
Replaces external vector DB tier
06 — Global enterprise

Workers at 330+ POPs for NEOM-style deployments

NEOM is in Saudi Arabia. Wells Fargo is US-only. Hitachi Vantara is Japan-headquartered. Wolters Kluwer is Dutch. Each Scry AI customer has different data residency requirements. Workers run at 330+ POPs globally — including Riyadh, Jeddah, Frankfurt, Tokyo — with per-customer regional pinning available. The "S" in IQ-SMART, geographically.

Workers Smart Placement Regional Services
Native data-residency compliance
07 — Orchestration

Workflows for the Collatio reconciliation pipeline

"Reconciles Instantly — Cross-document checks and decision-ready reports, automated end-to-end" — that's a multi-step durable workflow. Ingest → extract → cross-doc match → validate → annotate → approve → archive. Cloudflare Workflows is durable execution for this shape, with checkpoints, retries, and per-customer policy gates. Replaces Temporal or hand-rolled state machines.

Workflows Queues Cron Triggers
No external orchestrator to operate
08 — Concentio CCTV

Workers AI vision models at the edge for SceneTrack

SceneTrack runs CCTV analytics for safety, risks, and urban intelligence. Today vision inference probably happens regionally on GPU clusters. Workers AI runs vision models (CLIP, object detection, OCR variants) at the same POP as the camera — Riyadh cameras get Riyadh inference, sub-100ms per frame. Especially compelling for NEOM-scale smart-city deployments.

Workers AI Vectorize
Vision inference at sensor latency
Mapping

Scry AI capabilities → Cloudflare primitives.

Each Scry AI product line maps to specific Cloudflare developer primitives. Not approximately — exactly.

Scry AI capability What it does Cloudflare primitive
Per-customer enterprise isolation Each Fortune 500 customer = isolated config, corpus, policies Workers for Platforms dispatch namespaces
Collatio doc ingest + extraction Any-format ingest: PDFs, scans, tables, charts, schematics Workers + Workers AI (OCR + vision)
Collatio reconciliation workflow Cross-document checks, decision-ready reports, end-to-end Workflows + Queues + Durable Objects
Auriga multi-LLM conversational AI Chat + voice + avatar, multilingual, with source traceability AI Gateway + Workers AI multi-provider
Auriga Enterprise Knowledge Agent Governed RAG over enterprise data with source linking Vectorize + R2 + Workers AI Embeddings
Concentio digital-twin modeling Real-time stateful twin per IoT asset (camera, drone, meter) Durable Objects (1 DO per asset)
Concentio SceneTrack CCTV Visual intelligence from scenes — vision inference Workers AI (vision models at edge)
Datatio legacy modernization Reverse-engineer COBOL, PL/I — 60+ algorithms Workers for algorithm execution + R2 for artifact storage
Document corpus archive Customer doc history, audit trails, compliance evidence R2 (zero egress, S3-compatible)
Global enterprise deployment NEOM (KSA), Hitachi (JP), Wolters Kluwer (NL), Cisco (US) Workers at 330+ POPs + Regional Services
Quantify it

The AI Gateway math for Auriga across Fortune 500 customers.

Drag the sliders. The compounding insight: when N enterprise customers ask similar questions of their data, semantic caching scales with N. "What's our quarterly revenue trend?" "Show me last month's compliance exceptions" — these patterns repeat across customers in a domain.

Auriga AI Gateway savings calculator

Annual LLM inference cost — with and without semantic cache

Cache hits cost ~5% of a full inference call (embedding lookup + small response stitch). Adjust sliders for your actual scale.

50
5,000
2,800
45%
$15
Total Auriga calls / year 91M
Total tokens / year 256B
Cost without AI Gateway $3.8M
Cost with semantic cache $2.3M
Annual savings $1.5M

Directional. AI Gateway also adds free observability, rate limiting, fallback routing, per-customer cost attribution, and request logging — none of which is priced into the chart above. The compounding effect: as Scry AI adds enterprise logos, cache-hit rate goes up, not down.

Architecture

How a Wells Fargo loan officer queries Auriga on Cloudflare.

A loan officer in Charlotte asks Auriga: "Show me debt-to-income trends for my Q3 commercial real estate portfolio, flag exceptions." Following the full path.

1

Query hits the nearest Cloudflare POP (Atlanta)

The loan officer's Auriga client (chat/voice/avatar) sends the natural-language query to auriga.wellsfargo.scryai.com, which resolves to the closest POP — Atlanta. Round-trip time drops from ~95ms to ~14ms.

Workers Smart Placement
2

Workers for Platforms routes to Wells Fargo's namespace

Hostname → dispatch namespace lookup. Wells Fargo's worker — with their specific Auriga config, CognitiveBricks selection, RAG corpus pointer, and compliance policies — runs in an isolated runtime. Zero noisy-neighbor risk between WFC and Cisco using the same Auriga platform.

Workers for Platforms Dispatch Namespaces
3

Vectorize retrieves the loan-officer's portfolio context

Semantic search over Wells Fargo's isolated Vectorize index returns the top-K relevant loan records, debt-to-income definitions, commercial real estate policies. Per-customer index isolation — Cisco's policies never leak into WFC's results. Sub-30ms retrieval.

Vectorize R2
4

AI Gateway checks the semantic cache

The query + context fingerprint hits AI Gateway: "DTI trend analysis, CRE portfolio, Q3, exception flagging." Semantic search finds 23 similar resolved queries this quarter across WFC's loan officer pool. Cached response template + freshly-bound data returned. ~80ms.

AI Gateway Semantic Cache
5

If cache miss, route to the configured LLM

For WFC's compliance posture: route to Azure OpenAI Service (US East 2, data residency confirmed). Auriga's source-linking layer ensures every claim in the response points back to a specific loan record in the retrieved set. AI Gateway logs the full request + response for the WFC audit trail.

AI Gateway Azure OpenAI Logpush
6

CognitiveBricks for the structured analytics layer

The DTI trend analysis itself doesn't need an LLM — that's deterministic finance math. A CognitiveBrick (running as a Worker) computes the trend, identifies exceptions, prepares the visualization data. The LLM only narrates. This is your "quality, zero hallucinations" promise in primitive form.

Workers D1
7

Response streamed back via WebSocket

Auriga's chat/voice/avatar UI receives the structured response + LLM narration via WebSocket — token by token for the narration, full payload for the visualization. Sub-second perceived latency end-to-end.

WebSockets Workers
8

Audit trail archived to R2 for WFC's compliance

Full event trace — query, retrieved context, LLM call, source links, response, user identity — written to R2 (zero egress when WFC's auditors later request it for FFIEC examinations). The "Quality" and "AI-Powered" IQ-SMART covenants get their evidence chain.

R2 Logpush

Let's talk about the next 50 enterprise logos.

Scry AI's customer base — Microsoft, NEOM, Cisco, Wells Fargo, Kaiser Permanente, MassMutual — is the kind of list where every new logo doubles the infrastructure thoughtfulness required. The next tier of growth is where the platform-layer decisions made today compound for the next decade. A 30-minute architecture conversation, no slides, no sales pitch — just the engineering math and a whiteboard.

Book 30 min with Matt Holscher
Matt Holscher · Solutions Engineer · Cloudflare Developer Platform