Purpul Pulse

Backend structure and integrations,
without the boilerplate tax.

Describe the shape of the system you need. Get a working backend, an API, and a front end that talks to it. Ship when ready.

43x
lower running cost
from live billing
400 MB
memory footprint
against 15 GB
0
tokens per request
no model in the loop
100%
works offline
trading continues

Who is behind this

5 months of isolation.

That is the honest starting fact. The work began in early 2026 and was not demonstrated to anyone for five months, because of a test set at the beginning and not moved since:

“Can it do what it says, and will it genuinely make me more productive or help me? And my answer has to be yes, otherwise I do not build it.”

For five months the answer was not yes. That is why this is late, and it is also why the numbers in this deck are measured rather than projected.

What is shipped

A multi site trade counter and point of sale, built with a partner against their real requirements. The backend platform underneath it. A browser extension in evaluation.

What is not

The public product is in evaluation, not general release. Several claims in this deck are proven on our own machines and not yet at customer scale, and those are named where they appear.

How it is built

Small team. No outside investment. Every claim carries the measurement that produced it, and anything unproven says so.

The problem

Almost every backend is the same backend.

Tables, records, roles, permissions, search, media, reports, audit, exports, webhooks. Teams rebuild that same foundation for every project, then bill for it, then maintain it forever. Only a thin slice is ever specific to the business.

Rebuilt every time. Roughly 90 percent. Your 10 Where the budget goes Where the value is
The repeated part is the boilerplate tax. It is paid on every project, by every team, forever.

What changes

Describe it once. The foundation is already there.

The foundation is not written again for your project. It is already built and proven. Your requirements decide the shape it takes, and what comes out is a real backend with real endpoints, not a prototype.

You bring

Requirements, a spec, existing documents, or an existing system. Whatever you already have.

You get

A working backend, a documented API, an admin surface, and a front end wired to it.

You keep

The code, the data, and the ability to run the whole thing yourself.

Where to start

Write an RFP to yourself.

You already know how to do this. Describe what you need as though you were going out to tender: the things you keep track of, who is allowed to see them, what has to happen when, and what you would call a failure. Write it for you, not for a supplier.

That document is the input. There is no separate discovery phase to buy first, and no technical translation step where the meaning quietly changes.

What you write

“We run six trade counters. Staff take orders at a till and account customers have their own pricing. A discount over ten percent needs a supervisor. At close of day each till is counted and any difference is recorded.”

What comes back

The things you named, the rules you set, the people who may act, and the reports you asked for. Running, with your branding, ready to be argued with and changed.

The words you would have put in a tender document are the words the system is built from. If a requirement is missing from your description, it is missing from what comes back, and it says so rather than inventing one.

Running cost, measured

Two systems. Same infrastructure. One month.

Our platform Conventional stack
Our platform Conventional $1.39 $60.42 USD
Projected month. Ours was carrying active development traffic. Theirs was idle with no public use.
Our platform Conventional 400 MB 15 GB Memory
Same allocated processor on both. The difference is what the software needs to hold.
BilledOur platformConventional stackRatio
Month to date$0.12$5.0042x
Projected month$1.39$60.4243x
Memory400 MB15 GB37x

Figures read from the hosting provider's own billing for two live projects on the same plan.

Two ways to run it

One platform. A private deployment, or the browser.

For organisations

Deployed inside your estate, behind your own boundary. Your data stays where your policy says it stays. Custom front end where you need one, standard surfaces where you do not.

  • Runs on your infrastructure
  • Keeps working with no internet
  • Connects to systems you already run

For everyone

A browser extension. It makes sites easier to read and use, saves what you are browsing so you can work without a connection, and answers questions about the page you are on.

  • Accessibility built in, not bolted on
  • Offline reading and saved pages
  • Answers with no model and no tokens

Resilience

The line drops. The tills keep taking money.

Connectivity is treated as something that comes and goes, because it does. Work continues locally and is reconciled once the connection returns. Staff and customers are not asked to wait.

Connected Connection lost. Trading continues. Reconnected. Caught up. Sales recorded Sales still recorded, held locally Nothing lost, nothing re-keyed
Transactions taken while offline are held and settled automatically once the connection is available.

It watches itself

Quiet means checked. Not unattended.

One system watches everything you run: the cloud services, the servers, the machines on site. It stays silent while nothing has changed, and tells you the moment something starts to drift. Silence is a measurement, not an absence.

Silent Measured, unchanged, and therefore trusted. Speaking up Something moved. It says what, and when, by name. Cannot tell A reading is missing, and it admits that plainly. The third one is what makes the first one worth anything. A stale reading is never shown as a healthy one.
Most monitoring tells you when it is unhappy. The harder problem is knowing that quiet was earned.

Integration

It meets the systems you already have.

Most projects stall on the joins: the old database, the supplier feed, the payment provider, the warehouse. That work is normally the majority of the timeline and most of the risk.

Existing data

Brought across without a rewrite of the systems that hold it.

Third parties

Ordering, delivery, payments and logistics providers.

Identity

Works with the sign in and permissions model you already use.

Storage

Files and media, wherever you keep them today.

The aim is a shorter integration phase, not a promise that integration disappears. Scope is agreed per system.

Proof

Three claims. Run them yourself, offline.

We are collaborating with multiple partners on large scale enterprise products in completely different industries. Rather than ask you to take that on trust: three claims, each one a single command on your own machine, with no network, and each with a deliberate failure watched before the pass was believed.

1,143
source files
lifted to 530 entities
0.25s
to do it
on a laptop, offline
0
bytes of drift
across runs, days, rebuilt binary
3
reds
watched before any green

Why it holds up

Small, predictable, and honest about what it does not know.

Predictable

The same request gives the same answer. Behaviour can be tested, reviewed and signed off.

Small

A fraction of the memory and cost of a conventional stack, which is why it can run on modest hardware and in the browser.

Honest

When something is genuinely unknown it says so, by name. It does not invent an answer to look capable.

That last point is a design rule, not a limitation. A confident wrong answer costs more than a gap, because a gap announces itself.

In short

43x

Lower running cost. Same work.

A backend, an API and a front end from what you can already describe. Running for pennies, working offline, and connecting to what you already have.

See it

A working trade counter, the workspace behind it, and the billing that produced the numbers on this page.

Try it

Bring one system you would rather not rebuild. We will show you what comes back.