Cortex · Support Intelligence Engine

Support that pays for itself.

The self-hosted support intelligence engine for engineering-led manufacturers — grounded in your products, running on infrastructure you control, with every answer traceable to source.

Cortex case workspace: a magnetron-arcing case on a VaporX T15, with Cortex’s diagnosis, a five-step resolution plan and the four knowledge sources it cites, beside the engineer’s case list.
Cyber Essentials DEFCON 658 ready ISO 27001 aligned EU AI Act · Not high-risk ↓
97.2%
Citation rate
Locked 120-case benchmark
<1%
Hallucination rate
Locked 120-case benchmark
87.9%
Right source first
Held-out tickets, graded blind ↓
0
External AI calls
On every deployment
The problem we built Cortex to kill

Support is a quiet, compounding tax — £40k–£100k a year.

For an engineering-led manufacturer, customer support bleeds money in four directions at once, and none of it shows on a single P&L line.

Unbilled engineering

£40k–£100k of senior engineering time given away every year — untracked, unbilled, invisible.

Engineers stuck in triage

Your best people answer tickets instead of building product — and the knowledge walks out the day they leave.

Data as liability

The moment support data hits a cloud AI platform, you may already be in breach of an NDA or export obligation.

No pattern visibility

The same part fails across eight systems over six months — and nobody connects it until it's a warranty crisis.

Not a helpdesk. Not a chatbot. One platform, end to end.

From first contact to resolution — and it learns.

Cortex captures every case, resolves the routine work automatically, enforces your commercial policy, and turns recurring failures into product intelligence — learning a little more about your equipment with every case it touches. Self-hosted by default. Grounded in your data. Yours forever.

Capture & triage

Cases arrive from any channel into one record, then are scored on contract, severity, warranty and repeat-failure signals and assigned automatically.

Resolve

Cortex finds how similar cases were resolved before, proposes a grounded resolution, enforces commercial policy, drafts the reply, and raises a quote when free support is exceeded — for your engineer to review and approve.

Detect & compound

Recurring failures become R&D intelligence; every closed case feeds the knowledge base, so accuracy compounds with every ticket.

See it run

Twenty minutes on real data tells you everything.

A 90-second walkthrough of the case lifecycle, the conversion email, and the learning loop.

Capabilities — built and shipping

What makes it different.

01

Support that bills itself

The commercial policy engine tracks support time against each customer’s free-hours allowance, contract tier and warranty, detects the moment a case becomes billable, and drafts a three-option offer with the price worked out — for your engineer to review and send. Support stops leaking money and starts generating it.

Underneath: rate cards and quotes, with the allowance set per year. The Financials tab shows earned contract revenue and labour cost per role and team, alongside the cost of quality.

A three-option service offer Cortex drafted when a case ran past its free-support allowance: 6.57 hours estimated, 2 hours free, about £686 to pay. It is marked Draft · not sent, waiting for an engineer to review and send it.
02

Pattern detection → R&D intelligence

Recurring failures are clustered across unit, part and component, then raised as structured Product Improvement Recommendations for engineering — with the at-risk installed-base list attached. Reactive support becomes proactive engineering.

A pattern alert clustering a recurring failure across the installed base, with the drafted Product Improvement Recommendation
03

Equipment-aware knowledge

Knowledge is indexed by your bill of materials, not flat document storage. A case about one product pulls the right answer at product, sub-assembly and component level — specific to the exact configuration in front of the customer.

One installed unit in Cortex: its parts list on the left, and on the right everything known about one part — the bearing BRG-4410 — across the fleet: 11 known issues, 3 past resolutions and 10 documents.
04

A glass box, not a black box

Every recommendation cites its source; if it can't, it flags inference rather than asserting fact. Every parameter is yours to set — and the AI model is swappable, so there's no lock-in at the AI layer.

A Cortex diagnosis for a turbo-pump vibration case: the root cause citing the knowledge-base articles it relies on, a seven-step resolution rated High confidence, and the five sources behind it.
05

Just ask Cortex

A conversational assistant over your whole operation. Ask how to fix a fault and it answers from your manuals and past cases, with every source linked. Ask how to do something in Cortex and it walks you through it — that's your onboarding, no manual required. Ask who owns a customer's open cases, or this month's satisfaction score, and it answers from your live data, showing each person only what they're allowed to see. Sharp, permission-aware, and with a dry sense of humour.

It now draws on your own library of policies and procedures, keeps your chat history, takes dictation into any long field, and transcribes long voice memos.

Ask Cortex answering ‘What causes magnetron arcing on a VaporX T15, and how do we fix it?’: four likely causes, the fix in order, when to escalate, and the knowledge-base sources behind it — with a follow-up question underneath.
06

Works in the field — even off-grid

A native mobile app puts the whole operation in your engineer's pocket: update a case from a plant floor, capture a voice field report that transcribes itself, and read your cases even with no signal at all. On your own infrastructure, over a secure tunnel — because the work doesn't always happen at a desk.

The Cortex app on a phone in airplane mode: ‘Offline — showing cached data’, with the case and Cortex’s full diagnosis still readable.
07

Support milestones and the team, on one calendar

Case follow-ups and SLA due dates sit next to site visits, installations, on-call rotas and annual leave — so you see the work and who is free to do it in one view. Book leave with a named cover and new cases skip whoever is away. Open any day for the detail: which follow-up is due, at what stage, for which customer.

Cortex’s team calendar for November: case follow-ups next to site visits, installations, on-call, leave and company events, with one day opened to show a follow-up’s stage and a technical visit’s customer.
08

Know what support costs — and what it earns

Warranty cost by product line and by month, support-contract revenue and margin, and the labour nobody billed — worked out from the time your engineers log, at each person’s cost. It is one of 13 dashboards, and every chart exports to CSV or Excel.

Cortex’s Financials dashboard: warranty cost, contract revenue, contract labour cost, contract margin and £31,138 of unbilled labour over the last 12 months, with warranty cost by product line and by month.
09

SLAs that respect your working week

Every case carries a live first-response clock, counted in your working hours — closed days and bank holidays are skipped, and round-the-clock contract tiers run 24/7. The first reply settles the case as Met or Missed for good, so your compliance figures can’t be rewritten later.

An engineer’s case queue in Cortex with the live SLA clock on every row: two cases breached, three on track with hours left, three met.
10

Fits the stack you already run

No rip-and-replace. Cortex pulls your customers and installed base from the systems you already run: Salesforce, HubSpot, Zoho or Microsoft Dynamics for CRM; SAP, SAP Business One (file import), Priority, 123Insight or Zoho Books for ERP, or a plain Excel sheet if that is where your installed base lives. Tickets come in from Jira, Zendesk or Freshdesk, or straight from your Microsoft 365 support mailbox; documents from SharePoint or a network file share. Sign-on rides your own identity provider, and a Zoho CRM case link puts every Cortex case on the customer’s account. It reads from your systems of record, and writes only where you switch it on — which is exactly what your security team wants to hear.

Plus any SAML / OIDC provider · REST API & webhooks

Hover or tap a system to see what Cortex takes from it. Drag to turn the stack.

Shipped since July · in production

Built for the way your support team actually works.

Spam that never becomes a case

Unsolicited mail is dismissed without opening a case. A known sender is never dismissed automatically.

Security, deeper

An append-only audit trail with personal data redacted, documents read in a sandbox, and GDPR erasure that a later CRM sync cannot undo.

An appliance that looks after itself

A guided backup and recovery wizard, and a GPU that recovers by itself after a fault.

Value that compounds

Most platforms peak on day one. Cortex peaks on day one thousand.

Cortex learns from two streams — your company knowledge and your team's corrections. Every closed ticket becomes knowledge Cortex can retrieve and use; every expert's instinct becomes structured, reusable capability before it walks out the door. Where every other platform plateaus, Cortex compounds.

The one line to remember
Deal with it once. Learnt for life.
Defensible performance

Published numbers. Measured, not asserted.

Cortex publishes its accuracy, measured on the hardware we ship. Answer quality is scored on a locked, 120-case hand-authored benchmark; retrieval on held-out tickets the system has never seen. The full methodology is available on request.

  • A citation shows where an answer came from, and the engineer who reviews it decides whether it is right.
  • Clears NVIDIA's published 2026 production-RAG baseline of 64.8% — self-hosted, zero external AI calls [external figure, re-checked 10 September 2026].
  • In production at a UK scientific-equipment manufacturer.
MetricResultWhat it means
Answer quality · locked 120-case benchmark
Citation rate97.2%Cites the knowledge-base context the answer was built from
Hallucination rate<1%How rarely it asserts what it can't ground
Retrieval · held-out tickets
Measured on held-out tickets never used for tuning, graded blind by two independent AI reviewers, with the decision rules fixed before the measurement.
Hit@187.9%The right source ranked first
Hit@3100%The right source in the top three
MRR0.94Mean reciprocal rank — 1.0 would mean the right source always comes first
Speed & sovereignty
Diagnosis & resolution<60sFrom a new case to a diagnosis and a cited, proposed resolution
AI assistant response<2sTime for Ask Cortex to answer
External AI calls0Every inference runs on the Cortex server
Scale1,000Cases validated with sustained throughput and no degradation
An engineering claim, not a marketing one
We publish the table. The changes that didn’t work are in our measurement notes.
Sovereign by design

Your data, your AI, your walls.

Enterprise software with a formal security architecture — 30,000+ automated tests on every code change and a production-approved security risk score.

  • Data sovereignty. AI inference on single-tenant infrastructure using open-weight models — no OpenAI, Anthropic, Google, Azure or Cohere. Outbound AI calls blocked at the network layer and logged. Fully air-gapped on the Cortex Appliance.
  • Encryption. AES-256-GCM at rest, TLS 1.3 in transit — keys in environment, never in code.
  • Access & identity. SSO via SAML 2.0 and OIDC (Entra ID, Okta, Google, Ping), MFA (QR enrolment for admins), granular RBAC — 17 built-in roles plus your own custom roles, about 80 permissions.
  • Audit & oversight. Every action logged immutably for seven years; no AI-written email reaches a customer until a person sends it; every automatic AI effect is reversible and audited.
FrameworkHow Cortex aligns
GDPRSingle-tenant infrastructure. UK or EU residency, or the region you choose in your own cloud account. One-click erasure with full cascade, which a later CRM sync cannot undo.
ISO 27001Aligned to ISO 27001 controls across the stack.
DEFCON 658 · CSMv4Appliance fully air-gapped — no telemetry, all local.
Cyber EssentialsCertified.
EU AI ActAssessed not high-risk under Article 6. Article 50 transparency and Article 4 AI literacy met; source citation and human override throughout.
EU AI Act · Assessed 4 November 2026

Assessed against the EU AI Act — not high-risk, obligations met.

Cortex has been assessed against Regulation (EU) 2024/1689, as amended in 2026, and meets the obligations that apply to it. All AI runs on open-weight models on the server Cortex is installed on; no AI-written email reaches a customer until a person reviews and sends it; every automatic AI effect is reversible and audited.

  • Article 6 · Not high-risk — no Annex III use
  • Article 50 · AI-written email marked and sent by a person
  • Article 4 · AI-literacy manual with every deployment
Read the EU AI Act statement →

Self-assessed against the regulation's text and signed by our CEO. The Act sets no certification for systems that are not high-risk.

Three ways to deploy · one platform

Deploy it your way — and own it.

Appliance

On-premises · perpetual
  • Fully air-gapped, customer-owned hardware
  • Defence, NDA-heavy & regulated work
  • No internet required
  • Backup and recovery wizard; self-recovering GPU

Ascentis Private Compute

Dedicated hardware · in-country
  • A machine we run for you, in your country
  • SMEs wanting full data control
  • Funded and operated by Ascentis

Cortex Managed

Hosted · subscription
  • Fastest time-to-value
  • Teams without hardware, or cloud you already run
  • Single-tenant cloud: ours in the UK or EU, or your own AWS, Azure or GCP account

The same software and guarantees on all three options, with full data portability between them.

What it's worth

It pays for itself in year one.

Cortex is priced against what it returns, not what it costs. A typical client recovers more in unbilled support in the first year than the platform, implementation and maintenance combined.

£40–100k

Unbilled support typically recovered per year.

60–75%

Lower 5-year cost of ownership vs enterprise alternatives.

2–8 wks

To implement — not the 6–18 months of enterprise platforms.

Cortex · ROI & pricing

See what Cortex returns — and what it costs.

Set your team size and support operation. Your projected return updates as you go — enter a work email to reveal your indicative investment and our commitment.

Your support operation

10 seats
Everyone who'll use Cortex needs a seat — support and field engineers, their supervisors and managers, admins and leadership, plus other roles across sales, engineering, quality and finance. Minimum 3.
500
Across all channels — email, portal, phone, field.
20%
Out-of-warranty or beyond the free allowance — work you don't currently charge for.
£600
Billable repair, chargeable support, or conversion to a contract.
30 min
Average hands-on time before resolution.
£75/hr
Salary, on-costs and overhead — what an engineer-hour really costs.

Your year-one return

Support revenue recovered
£60,000
Billable work you stop giving away — and convert instead.
Engineering time freed
175 hrs (~£13,125)
Senior hours handed back to product development.
Reveal your indicative investment & ROI

A valid work email is required to reveal your pricing. We'll send you a copy and follow up to talk it through — nothing else.

Please enter a valid email address.

Your return
On your numbers, Cortex pays for itself in around 4 months.
Indicative investment — from £15,480 / year
Our commitment

We make it work. If Cortex doesn't deliver against the plan we agree, we don't hand back cash and walk away — we keep going: extended support, further training and continued optimisation, until it does.

Your figures are on their way — let's walk through them. A short call to confirm the tailored figure for your deployment and put the commitment in writing.

Book a call →
How this is calculated. Return — revenue recovered = cases × recoverable % × average value; engineering time freed assumes Cortex resolves or assists ~70% of routine cases without an engineer (conservative: Cortex puts the right source first on 87.9% of held-out tickets), valued at your hourly cost. Investment — Professional licence (Core Platform + the Commercial Policy Engine that drives the recovery above + Pattern Detection) at your seat count; seats 16+ are billed at 15% off. Licence only: one-time deployment and any hosting are quoted by deployment model, and annual prepay (−20%) plus the first 60 days free reduce Year 1 further — all confirmed on the call. The return shown reflects support recovery only — seats in other roles (engineering, sales, quality, finance) add value beyond what's modelled here. Figures are indicative, not a quotation.

Appliance

Perpetual licence
One-time licence + annual maintenance
  • Own it outright, on your hardware
  • Year-1 maintenance included

Ascentis Private Compute

Perpetual licence
One-time licence + managed hosting
  • A machine we run for you, in your country
  • Funded and operated by Ascentis — you own the licence

Cortex Managed

Subscription
Per seat · monthly or annual
  • All recovery modules included
  • Fastest start, lowest entry cost
Why Cortex

Every other AI platform sends your data out. Cortex keeps everything in.

Cloud helpdesks have no hardware lifecycle and are ruled out where data can't leave the building. Enterprise field-service platforms are priced for £500m-plus companies. Cortex is built for the engineering manufacturer who needs working support intelligence, owns the result, and refuses to send their product data anywhere.

Resources

Take it away with you.

📄

Product brochure

The 8-page overview — what Cortex does, the proof, deployment and security.

Download PDF →
📋

Datasheet

The full public datasheet — capabilities, benchmarks and compliance.

Download PDF →
▶

90-second demo

Watch the case lifecycle, the conversion email and the learning loop.

Watch the demo →
🔒

Security & Data Protection Overview · gated · NDA

For your compliance team — architecture, controls and the compliance map.

Request access →
📊

Benchmark methodology · on request

The full, reproducible measurement spec behind the published numbers.

Request the spec →
🧮

ROI estimate

See what Cortex could recover for your support operation.

Estimate your ROI →
⚖️

EU AI Act statement · New

How Cortex was assessed under the EU AI Act: not high-risk, with its transparency and AI-literacy obligations met. Signed and dated, three pages.

Download PDF →
Questions engineers ask

Straight answers.

What about our data?

On the Cortex Appliance it stays on your hardware. On every option your data sits on single-tenant infrastructure with UK or EU residency and is never sent to an external AI provider. On Ascentis Private Compute it sits on a dedicated machine we operate in your country. On Cortex Managed it runs on single-tenant cloud, ours in the UK or EU or your own AWS, Azure or GCP account. On every option there are no external AI APIs, your data is never co-mingled with other customers, and you own the data and the knowledge base it builds.

What if the AI gets it wrong?

Every suggestion is reviewed by a human before it reaches a customer — Cortex drafts, your engineers approve. Hallucination is under 1%, and 97.2% of recommendations on tickets with a documented answer cite a real source. Every correction improves the knowledge base where the problem actually is.

What happens when a better AI model comes out?

That's your advantage, not your problem. Cortex is model-agnostic — you pull a better model, load it from the admin UI, benchmark it against your own corpus, and switch. No code changes, no upgrade fee, no lock-in at the AI layer.

We're a small team — is it overkill?

If two or more engineers lose time to things that aren't R&D, you're the right size. Smaller teams benefit more, because tribal knowledge is a bigger risk.

How long does implementation take?

Two to eight weeks depending on data migration. You're not buying a blank canvas — Cortex ships with a proven support workflow already configured.

Does it work for our specialist equipment?

It learns your products during onboarding — from your manuals, historical tickets and a knowledge session with your engineers. The value is the knowledge base it builds about your equipment, and it keeps learning every day.

We've always given free support — does Cortex make us charge?

No. Cortex is policy-agnostic: if free support is your strategy, it makes free support 25–30% cheaper to give. Where a case is already chargeable under the policy you run today — out-of-warranty work, spares visits — it catches the moment and drafts the offer. Whether it gets sent stays entirely your call.

Won't raising a charge annoy our customers?

Engineers would rather fix things than talk about money — so Cortex does the talking. It drafts a polite note offering three options: a one-off fix, a support pack, or a contract. A choice reads very differently from an invoice, and a human always presses send. And in practice, nobody re-tenders a £250,000 instrument to avoid a £600 quote; what loses customers is the case that stays open for months.

See it for yourself

Twenty minutes on real data is all it takes.

You'll know within those twenty minutes whether Cortex solves a problem you have.

Pick a time that suits you

On Microsoft Teams. Bring a real support case and we'll run it through Cortex live.

Book your call →
Or start with the two-day Diagnostic

We profile your support data and the systems around it, then hand you a Data Health Report and an implementation plan: what Cortex will run on, what to fix first, and when it goes live.

Ask about the Diagnostic →

Accelerate. Optimise. Outperform.

Weighing local inference against a cloud API? Our 2026 self-hosted AI build, benchmark and cost guide sets out the real numbers.