Third-Party Software

In short

Our systems are built from a mix of open-source components, commercial services, and software we write ourselves. We meet every licence obligation attached to third-party components, and the corresponding notices are included in every handover of code to a client.

What we build, and what we do not

We do not train foundation models, and we do not claim to. Companies that claim otherwise almost never do, and it is easy to check.

What we build is the layer that decides whether an agent is fit to be trusted with your company's data:

• Written limits of authority — what the agent may do unasked, and what must wait for a person

to approve.

• An audit trail for every decision — what it read, what it decided, and on what basis.
• Fallback providers — when the primary provider goes down or runs out of quota, the work moves

rather than stops.

• Measurement of right and wrong — quality measured against cases whose answers are already

known, not judged by impression.

• Defences against injected instructions — outside text is treated as data, never as commands.
• A cost ceiling per job — the cost of every run is recorded and capped.

That layer is the actual work. The part anyone can obtain is not the part that makes a system last.

The engine is replaceable, deliberately

Our systems are not locked to one provider or one model. Either can be swapped from a panel without rebuilding anything.

The reason is not tidiness but risk control. AI engines change their prices, change their terms of use, and stop being maintained — several have done all three in the past two years. When one of them changes, your service does not change with it. What persists is the layer above, and that layer is yours the moment it is handed over.

A consequence other suppliers rarely mention: because the engine is replaceable, we have no interest in defending any particular vendor. If a cheaper one produces the same result, the saving is yours.

Due diligence

Your legal, procurement, or security team will usually need a component inventory with licences before signing. We prepare that inventory in full and give it directly to your team, under a non-disclosure agreement where required.

We do not publish it. The detail of how a system is assembled is useful mainly to parties with no stake in it, and publishing it does nothing for our clients.

Request it at hello@neuraltan.com.

The code you receive

Every handover includes source code, documentation, and the third-party licence notices that go with it. Nothing is locked away, and nothing requires you to keep us on in order to run it.