Technical interface / Deployment design
A familiar interface.
A controlled deployment.
Our interface design uses an OpenAI-compatible request structure, with identity, models, data handling and service requirements defined for each deployment.
The application interface
Our interface design connects applications to controlled intelligence infrastructure. The deployment defines identity, permitted models, data handling, metering and service requirements together.
Interface reference — deployment endpoint and credentials supplied separately.
# DOCUMENTATION SHAPE — NOT A LIVE OR PUBLIC ENDPOINT
POST <deployment-base-url-if-agreed>/v1/chat/completions
Authorization: <deployment-specific credential — not issued here>
Content-Type: application/json
{
"model": "<accepted-model-id>",
"messages": [{"role": "user", "content": "..."}],
"stream": "<accepted-setting>"
}Placeholders do not identify an available base URL, issued credential, accepted model or supported streaming mode. Browser-delivered reusable credentials are not an approved pattern.
Design the deployment around the workload.
Define the configuration, operating agreement and acceptance evidence together.
- Interface and identity — service name, base URL, client identity, credential delivery, scope, rotation and revocation.
- Workload — accepted model and configuration, precision, context, latency and useful-output acceptance test.
- Residency — permitted execution and data locations, named explicitly.
- Data — retention, logging, prompt handling and deletion behaviour.
- Security — trust chain, authentication and authorisation, key management, review and supplied evidence.
- Availability — accepted failure set, recovery objective and support window; no SLA exists until contracted and evidenced.
- Metering — facility input, accelerator time and useful-output measure.
- Evidence — telemetry and attestation artefacts supplied to the customer.