Role
For CTOs
One gateway, many models. Switching model changes a setting, not your application code.
What is at stake here.
Provider lock-in
OpenAI, Gemini and local models sit behind the same gateway.
The latency budget
The inspection chain adds under 2 ms at p95; what determines end-to-end time is the model provider.
Can a team bypass it?
The order is fixed in the orchestrator; skipping a step from application code is not possible.
Why it is not optional
This debt comes back with interest.
Where a legal duty is the reason, the article is named. No penalty figures and no circular numbers — those change, and a marketing page is the wrong place to be wrong about them.
Distributed security is distributed failure
Five implementations of the same control in five services is five separate gaps. A shared layer reduces it to one surface.
Provider lock-in is architectural debt
Wiring clients straight to a provider turns the decision to change model into an engineering project.
The overhead has to be measurable
Any layer in the path adds latency. Publishing a measured figure for it is what makes the trade-off arguable rather than assumed.
Use cases
What actually happens in the day's work.
These are mechanisms, not case studies. Each one is a concrete thing somebody does, and what the chain does about it.
One gateway, several models
Situation
Different teams use different providers, each carrying its own key and its own error handling.
What happens
They all move behind the same OpenAI-compatible gateway. Clients do not change; only the base URL does.
Switching provider
Situation
Cost or performance calls for a different model, and you do not want to touch application code.
What happens
The model is a setting. OpenAI, Gemini and local models sit behind the same gateway.
Writing security into every application
Situation
If masking, policy and logging are implemented in every service, each one is implemented wrongly in its own way.
What happens
The chain runs in one place and every application gets the same decision. There is no security code to write for a new service.
The answer
Which part of the product responds.
Configuration
Recommended starting posture.
These are starting recommendations; the final policy is written with you during setup.
- Deployment
- SaaS, VPC or on-premise
The rules get written together
The rules for this sector are written with you during onboarding: blocked phrases, the competitor list and the entity rules are set against your own processes. Instead of adapting someone else's template, you start with a policy that is genuinely yours — and it goes live only after being tried in the simulation that runs without saving.
Other roles
Let's talk through your own scenario.
In a 30-minute technical session we run your prompts and your policy through it live.