Operations · South Africa

Splide builds and runs the systems businesses depend on.

That work spans infrastructure, operational software and the answering service behind 2D barcodes. Behind it, Splide builds and uses AI agents with persistent memory where they make the operation better.

See what we run How we work

Work

Built and operated

Memory

Organisational, not disposable

Your data

Always yours

Handover

There is none

splide/infraown it before it breaks

splide/buildpriced per system

splide/codeanswering for years

splide/infra

Infrastructure

Infrastructure for companies whose technology matters but nobody inside the business owns it. We take responsibility before something breaks, not after.

networks
servers
identity
backups
monitoring

splide/build

Business systems

Systems built around the way a business actually works, then operated as part of ours. Agents are used where context and judgment help. Deterministic work stays deterministic.

ordering
reporting
workflow
agents
operations

splide/code

2D barcodes

The answering service behind GS1 2D barcodes. We coordinate the pilot and keep the service behind the printed code answering for the life of the packaging.

product identity
approved data
resolver
verification
operation

Agents that survive the session

An AI model can do impressive work and still begin the next session knowing nothing about it. That is not enough for a system a business will depend on.

Splide's agents work with persistent memory. They can retain decisions, connect related facts, surface contradictions and carry operational context from one session to the next. The model can change without the organisation losing what it learned.

That memory does not make every decision correct and it does not remove human accountability. It removes the absurdity of intelligent systems repeatedly starting from zero.

See Adja

Adja

adja.ai

Persistent organisational memory for AI agents, and the memory platform Splide's own agents use. It recalls by relevance and relationship rather than keyword alone, allows old information to decay, consolidates related memories and surfaces contradictions instead of silently choosing one. The model is replaceable; the organisation's memory is not.

A semantic graph, not a document store
Spreading activation across linked memories
Salience, decay and reranking
Consolidation, so old memories are merged not lost
Contradictions surfaced, not silently overwritten
Isolated tenants, one per agent or project
Provider-agnostic, exposed over MCP

Nova

Operational tooling built to run real infrastructure. Nova holds the estate together and is the reason the infrastructure work is not manual. It is not a product and it is not for sale. It is evidence that the work is operated as a system, not as a pile of tickets.

Discovery and inventory
Encrypted credential store
Firewall, Proxmox, UniFi, Active Directory
Diagnostics

A distributor operating system

Built around a business, still running years later. A bicycle distributor runs warranty claims, service bookings and courier bookings on a platform Splide built. The client owns it outright. Splide still hosts, patches, monitors and changes it, with a reporting subscription running alongside it. The system is theirs. The operation is ours.

Warranty claims, service and courier bookings
Owned outright by the client
Hosted, patched and monitored by us
Reporting runs as a separate subscription

The Splide resolver

demo.splide.com

A public, GS1-conformant answering service. It runs against version 1.2.1 of the GS1-Conformant Resolver Standard and passes GS1's published conformance test suite. It is public because a checkable technical claim is worth more than an asserted credential.

GS1-Conformant Resolver Standard 1.2.1
Passes GS1's published conformance test suite
Answering at id.splide.com

Habistats

habistats.info

An environmental livability index, built because the data was worth having. Not a product. It is here because we still run it.

How we work

01
Understand the operation
A feature list is not a business. We start with the process, the people, the exceptions, the existing systems and the consequences when something goes wrong.
02
Build the right shape
That may require ordinary software, agents, persistent memory, infrastructure automation or all of them together. We do not force AI into work that a rule can do better.
03
Stay responsible
We do not disappear at handover. The system is monitored, maintained and changed as the business changes. The person learning from it is the person still accountable for it.

Who runs it

Jacques Karsten is the builder behind Splide. Over three decades, his work has moved from servers and networks through enterprise data and web platforms to operating software, agents and persistent memory.

The through-line is simple: understand an operation deeply, build what it actually needs, and remain close enough to know when reality changes.

Today that work is visible in Adja, Nova, the distributor platform and the public GS1-conformant resolver. The person making the technical judgment is also accountable for what happens after launch.

Why these belong in one company

Running a domain controller for a firm with no internal IT owner is the same kind of responsibility as running an ordering system for a distributor with no software team, or answering a product code years after the packaging was printed.

In each case, the business depends on something continuing to work and improve while nobody inside the business is going to operate it. Splide takes that responsibility and uses its own systems, agents and memory to do it with more continuity than a conventional support model allows.

The technology changes. The job does not.

The fee buys operations. It never buys permission. Stopping the fee stops Splide operating the system. It does not stop access to the system, the credentials or a single row of client data.

Build it around the business. Keep it alive as the business changes.

Talk about the system