Wants one enforcement point for all AI traffic, instead of policy that differs per tool.
Platform · Control plane
The control plane
The heart of Numaga: every AI interaction in your organisation runs through one central gate, where policy is enforced deterministically — before a model comes into play.
Who it is for
Wants personal data masked before the model call — demonstrably, not on trust.
Wants governance that is central, not reinvented for every new use case.
The problem
Governance after the fact is not governance.
In most organisations AI policy lives in documents and good intentions, not in the system. What an employee sends a model is barely auditable afterwards, and certainly not preventable.
The control plane inverts that: policy is code that runs before every model call. What does not pass the gate does not happen.
Control plane — the journey of a prompt
Six checkpoints, before a single token reaches a model.
Pick a prompt and follow it through the control plane. Every step runs before the model call — no compensation after the fact.
Example prompts
Prompt arrives· · ·
Analyse this patient record and propose a follow-up treatment.
Processing…
Optimised model
01
Classify
Sensitivity label assigned — public · internal · confidential · strict.
02
Redact
Special-category personal data masked before it leaves the perimeter — GDPR Art. 9.
03
Risk-gate
Checked against the EU AI Act risk class — prohibited use is blocked.
04
Route
The best-fitting model chosen internally based on risk and sensitivity.
05
Meter
Budgets and rate limits enforced per user and per team.
06
Log
Immutable audit record written — tamper-evident, time-synced.
Synchronous
Every step runs before the model call — no compensation after the fact.
Deterministic
Policy rules, not LLM judgement. Reproducible for inspectors.
Configurable
Routing tables, sensitivity rules and budgets tuned to your policy.
If it doesn't pass the control plane, it doesn't happen.
See the control plane in your own environment.
We set up a demo environment and run the six checkpoints against your own policy.
Get demo access