Web apps for the agentic era.
More visitors. More agents. Less infrastructure drama.
Global presence under 100 milliseconds.
Continuous evidence
inputs tested
Peachfuzz sends millions of malformed, unexpected, and adversarial inputs through Kernel.
It uncovers crashes and security flaws before they can weaken the platform every Numinos surface inherits.
Runs continuously. Every figure above comes from tamper evident, replayable run records.
Agent posture
A.I. stays off until the job asks for it. The public page gets speed first: fewer moving parts, quicker reads, cleaner delivery.
Enjoy the speed. Keep the surface simple.
Performance
Your app stack still matters. Blink gives it less public traffic to carry.
Target marker
1M
requests / second
Treat each row as one benchmark-class app server. Then the question is simple: how many boxes stay warm?
Server-month estimate
~$36
one always-warm 6.25M/sec box
Blink clears the 1M marker with one box. Slower rows buy more boxes to stand in the same place.
Returns compound
The page arrives sooner. The bill has less room to grow.
Throughput comparison
Same job: return the page. More throughput means each hosting dollar carries more visitors before scale-up becomes urgent.
Black marker = 1M RPS target. Server counts round up.
Go / Blink 1 server · ~$36/mo
6.25M/ sec
Java / Spring Boot 4 servers · ~$144/mo
279K/ sec
PHP / PHP-FPM 4 servers · ~$144/mo
278K/ sec
Python / FastAPI 8 servers · ~$288/mo
140K/ sec
Ruby / Rails 13 servers · ~$468/mo
81K/ sec
Business impact
Speed is not a vanity metric. It is product quality people feel before they can name it.
What changes
Bounce
53%
Blink helps the visitor meet the product before waiting becomes the memory.
Conversion
+8% to +10%
Intent cools down with every visible hitch between click and payment.
Friction
$260B
Delay gives visitors time to question the product, the price, and the company behind it.
Search
Fast pages are easier to crawl, easier to trust, and easier to keep fresh.
Owners
Relief without turning the roadmap into a rewrite story.
Repeat
A fast return visit quietly says the product is in order.
For users
For owners
The team gets more headroom, steadier tails, and fewer emergency architecture meetings.
Insights
People, crawlers, and agents all meet the same first impression.
Black Friday carts. Breaking-news readers. Chatbots checking sources. Different arrivals, same first question: did the page show up?
First impression test
Compare
Same arrival. Different aftertaste.
Slow public page
People: Wait. Doubt. Leave.
Blink public path
People: Arrive. Trust. Act.
Slow public page
Crawlers: Defer the page.
Blink public path
Crawlers: Index the page.
Slow public page
Agents: Skip the source.
Blink public path
Agents: Cite the source.
About Us
Numinos combines company-building, private-market sales, and low-level systems work. Blink exists because public-web delivery is now an operating cost leaders can feel.
CEO
Mike sold his first company, a document management business, to private equity for £50m in 2006. That sale helped him recognise there was a need for an independent advisory firm to help sell privately held companies. He subsequently founded Convex Capital to fill that gap. In 2019, Mike sold Convex Capital to a UK listed professional services company for £22.5m after having advised on the sale of over 100 private companies with a combined valuation of over £3billion.
Having sold his own businesses and advised on the sales of so many others Mike knows the biggest post-sale challenge facing entrepreneurs is working out what to do with the money. Here again, Mike recognised the need for an independent investment firm catering to the needs of entrepreneurs.
Mike became a client of Equitile, shortly after the firm launched in 2016. In 2026, after a decade as a client and after his initial investment with Equitile had tripled in value, he joined the company as Chief Sales Officer.
When not working Mike turns his hand to farming. His latest venture is a remote fruit farm which he is using to preserve endangered varieties of fruit trees, herbs and other traditional agricultural plants.
CTO
Ase Deliri is a full stack engineer and the creator of Blink, the delivery engine behind Numinos.
Before building Numinos, he worked in telecom engineering at Rogers Communications, where he developed an interest in latency, network behavior, system capacity, and the operational cost of inefficiency at scale.
His work focuses on Go, distributed systems, Google Cloud infrastructure, frontend delivery, low level performance engineering, security, and automated test systems. He tends to work closest to the parts of software where runtime behavior, network delivery, infrastructure, and user experience intersect.
Blink grew out of that approach. Measure real bottlenecks, reduce unnecessary runtime work, and build systems that can verify behavior directly instead of relying on assumptions. The goal is simple. Make web delivery faster, more predictable, and easier to reason about under load.
Transport
Marketing, product, and admin keep separate rooms without becoming separate stories.
Org Benefit
Teams get clear boundaries without buying stack drift, duplicated operations, and uneven reliability.
Switchboard
Website, webapp, and admin resolve at the routing layer. Shared ground stays shared.
Service Boundaries
Website stays lean. Product keeps tenant and workflow depth. Admin keeps its own control posture.
Deployment Split
Marketing can ship without waiting on admin. Admin can ship without freezing product.
Team Continuity
Leaders and engineers reason about the same platform even when their work lives on different surfaces.
Transport Flow
The gain is organizational as much as technical: fewer dependencies, fewer staged releases, and fewer local changes waking unrelated systems.
Signal
Split codebases.
Signal
Release coupling.
Signal
Context switching.
Signal
Transport drift.
Outcome
Shared platform. Clean isolation.
Tenants
Tenant boundaries stay tied to explicit platform scope.
Campaigns
Campaign surfaces can come and go without pulling the whole estate with them.
Change
Blast radius stays local. Refactors do not become platform-wide negotiations.
Outcome
Shared platform. Clean isolation.
Tenants
Tenant boundaries stay tied to explicit platform scope.
Campaigns
Campaign surfaces can come and go without pulling the whole estate with them.
Change
Blast radius stays local. Refactors do not become platform-wide negotiations.
Delivery
The public path no longer carries normal app-shaped cost. Response time and memory stay easy to read under load.
Lumen
The visitor receives a usable document immediately. Live state does not hold first paint hostage.
Ballast
Repeat visits stop buying the same payload again.
Live Data
Live updates stay attached to the parts that need them.
Chisel
Full hierarchy without making typography a vanity payload.
Waiter
Failure stays bounded instead of turning weak networks into user-facing chaos.
Business Effect
Lower cost per page view. More public capacity per compute dollar. More time before the next forced infrastructure meeting.
Page Delivery
Under 300 ns
The document path stays in nanosecond-class latency.
Asset Delivery
Under 300 ns
Static assets keep the same discipline.
Font Budget
137 KB total
Three families, full hierarchy, and a tight payload.
Security
Security belongs in the structure. Public routes, authenticated routes, admin surfaces, tenant boundaries, and runtime posture share one model.
Wall
Invalid traffic is rejected before feature code starts work.
Surface
Route intent is explicit, not guessed from naming.
Gate
Public and authenticated paths keep different contracts.
Admin
Missing admin authorization is a hard stop.
Fairness
Tenant limits protect shared service quality.
Route Contract
Public, protected, elevated, and admin are platform contracts. They do not live in tribal knowledge.
Proof
Posture, checks, and evidence stay queryable in the running system.
Catalog
Named checks with clear boundaries.
History
Audit lineage that survives production.
Public traffic, sessions, tenant boundaries, and admin control stay inside one security posture.
Boundary
Tenant isolation stays structural.
Session
Identity state changes are explicit and reversible.
Admin
The control plane fails closed when authorization is missing.
Traffic
Invalid traffic is rejected before it becomes application work.
Telemetry
When a route fails in the browser, the evidence stays connected. Client events, traces, logs, metrics, and alerts stay on one path.
Golden Thread
One identity across the incident.
Cyclops establishes the golden thread. Browser evidence comes back on the same path.
Photon
Performance evidence returns on the original trace.
Cyclops
Trace identity stays attached to the response.
Prism
Browser exceptions keep their request story.
Golden Thread
One identity across the incident.
Cyclops establishes the golden thread. Browser evidence comes back on the same path.
Outcome
A checkout failure can be followed from user action to backend fault.
Cyclops
Trace identity stays attached to the response.
Photon
Performance evidence returns on the original trace.
Prism
Browser exceptions keep their request story.
The gain is diagnostic speed. Support, product, and engineering can read the same path.
Client
Browser evidence stays attached to the same identity.
Server
Logs and metrics stay route-aware.
Support
A failed payment resolves to traceable evidence, not guesswork.
Ops
Alert volume reflects posture, not raw event count.
Infra
Payments, writes, storage, export, search presence, and release discipline stay inside one operating model. Infrastructure should widen choice, not narrow it.
Money Rails
More payment rails should not mean more payment stories.
Checkout
One commercial entry point with room for more rails.
Fulfillment
Payment behavior stays auditable and clean.
Write Posture
A wider data surface should add choices, not accidental choreography.
Single
One backend when simplicity is right.
The platform does not force choreography when one durable store is enough.
Dual
Parallel writes for staged migration.
Migration can proceed in steps instead of one expensive cutover.
Multi
Clear boundaries across a wider data surface.
As storage concerns widen, the operating model stays deliberate.
Backends
Contracts
Application code points at stable contracts.
Boundaries
External dependencies stay bounded enough to replace.
Backend choices should stay negotiable. Migration should feel operational, not existential.
Exodus
Customer data should remain exportable, verifiable, and transferable.
Integration posture stays inspectable.
Presence
Contacts, hours, and business details should power the product and stay legible to search engines and machine crawlers.
Search
Structured metadata stays part of delivery.
Release
Deployment posture stays deliberate.
Kernel
Shared ground supports new products without forks.
Result
More reuse without flattening the product.
Conclusion
The principle is optionality. Adopt the delivery layer alone, keep existing auth and payments, or widen scope only where the gain is real.
Take What Fits
Delivery, auth, payments, telemetry, export, and storage posture can arrive without an all-or-nothing migration.
No Trap
Start narrow. Expand where return is measured.
No Drift
Shared ground stays coherent as surfaces diverge.
Edge
The delivery layer can stand alone.
Identity
Existing identity can stay until migration earns its place.
Money
Payment posture stays modular.
Data
Storage, export, and migration paths stay open.
Platform Longevity
A smaller surface means updates feel like housekeeping, not negotiations with abandoned dependencies.
Upgrades
Maintenance, not negotiation.
Dependencies
Fewer stall reasons. Less upgrade drag.
Kernel
New products behave like distributions, not forks.