Technical Architecture

End-to-end eight-layer architecture spanning from customer applications down to physical GPU infrastructure, storage, and monitoring systems.

3. Architecture Overview

4. Infrastructure Information

Operational deployment details. Values marked in amber should be completed with actual operational records in production.

• Cloud Provider[Insert Actual Provider]
• Server Quantity[Insert Actual Quantity]
• GPU Model[Insert Actual GPU Model]
• GPU Quantity[Insert Actual GPU Quantity]
• Deployment Region[Insert Actual Region]
• SLA Guarantee99.9%
• Annual Token Quota (Agentic)212,000,000,000 (212B)
• Technical Support TierFull API Technical Support

API Documentation

Complete API reference for the PREBO AI Computing Platform. All endpoints live under the Base URL and use Bearer API Key authentication.

Base URL for all API requests. Append endpoints below to construct full URLs.

Authentication

Every API request must include an API Key in the Authorization header using the Bearer authentication scheme. API keys are provisioned when the customer account is created and activated — see User Manual → Quick Start.

Header
// Example Authorization header
Authorization: Bearer sk-prebo-xxxx-xxxxxxxx-xxxxxxxx
Content-Type: application/json

Inference Endpoint — /v1/inference

The primary inference endpoint — submit a model identifier and an input payload; receive a completed result, request ID, and status tag.

Perform inference against an available AI model.

Request Parameters

Request Example · cURL
# POST /v1/inference
curl https://api.prebo.fun/v1/inference \
  -H "Authorization: Bearer sk-prebo-rouse-ml-2026" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "gpt-4o",
  "input": "Explain the PREBO AI Computing Platform in one sentence.",
  "params": {
    "temperature": 0.7,
    "max_tokens": 512
  }
}'

Response · 200 OK
{
  "request_id": "req-8f3a-91b2c7d4",
  "status": "completed",
  "model": "gpt-4o",
  "result": "PREBO AI Computing Platform is an enterprise-grade AI infrastructure product by PREBO TRADING LIMITED that delivers dedicated GPU compute, end-to-end model training, standardised inference APIs, data processing, and an annual 212B-token Agentic AI package with 99.9% SLA.",
  "usage": {
    "prompt_tokens": 128,
    "completion_tokens": 64,
    "total_tokens": 192
  },
  "elapsed_ms": 842,
  "created_at": "2026-08-02T08:15:33Z"
}

Model Endpoints

Inspect which models are currently enabled on the platform.

List all available models with platform and capability metadata.

Retrieve details for a single model.

Response · 200 OK (list)
{
  "object": "list",
  "data": [
    {
      "id": "gpt-4o",
      "object": "model",
      "owned_by": "openai",
      "platform": "openai",
      "type": "chat",
      "context_window": 128000
    },
    ... more models
  ]
}

Training Service Endpoints

End-to-end model training job management. Matches the 5-step workflow in the Product Overview.

Create a new training job. Upload data first via POST /v1/training/data.

Check training progress, metrics (loss, accuracy), and current status.

Download the trained model artifact (weights + config) once the job reaches "succeeded".

Job Status Enumeration

Usage & Billing Endpoints

Token usage, platform breakdown, subscription status, and SLA reporting.

Account-level usage summary (KPI totals, today's snapshot, performance metrics).

Daily usage timeseries for the last N days (default 7).

Current subscription plan, quota usage %, SLA status, and period timings.

Error Codes

All errors follow the RFC 9457 / OpenAI-compatible shape with a nested error object.

Error Response Example
{
  "error": {
    "message": "Invalid API key provided.",
    "type": "invalid_request_error",
    "code": "invalid_api_key"
  }
}

User Manual

Step-by-step guide from account creation through training and production API integration.

Quick Start — Go Live in 5 Steps

1. Customer Account Creation

   PREBO TRADING LIMITED provisions your organisation account and records it in the service delivery log. You receive an activation email with your account ID and the contract reference.

   • Verify your email domain(s)
   • Set up at least one administrator account
   • Accept Terms of Service and Data Processing Agreement
2. API Credential Issuance

   Navigate to the Dashboard and generate your first API key. The key is displayed exactly once — store it securely.

   • Recommended: create separate keys for dev, staging, production
   • Apply least-privilege scope tags (inference-only, training-only, admin)
   • Rotate keys at least every 90 days
3. Service Activation

   Your Annual Agentic AI Dedicated Computing Package (212B Tokens, 99.9% SLA) is activated at the Service Activation Date recorded in your contract. Token counters start from zero on that date and run for 12 months.

   Example activation: Customer "ROUSE MICHAEL LLOYD" activated on 2026-02-18 with an Annual Agentic AI Package of 212B Tokens. Expiry date: 2027-02-18.
4. Send Your First Inference Request

   Python · First Request
   import requests
   
   HEADERS = {
       "Authorization": f"Bearer {API_KEY}",
       "Content-Type": "application/json",
   }
   resp = requests.post(
       "https://api.prebo.fun/v1/inference",
       headers=HEADERS,
       json={"model": "gpt-4o", "input": "Hello PREBO!"},
   )
   print(resp.json()["result"])
   

5. Monitor Usage on the Dashboard

   Check the Dashboard for remaining Token budget, platform-level breakdown, content-type distribution, and per-day trend charts. Configure email alerts at 80% and 95% of quota use.

How to Run a Model Training Job

The platform supports fine-tuning and custom training via the 5-step workflow below.

1. Upload Training Data

   Use POST /v1/training/data with a multipart upload, or the Dashboard's drag-and-drop uploader. Accepted formats: .jsonl, .parquet, .csv, .zip (images). Max 50 GB per dataset.

2. Configure Training Parameters

   Choose a base model, set hyperparameters (learning rate, batch size, epochs, warmup ratio), select the target GPU tier, and pick a validation split.

3. Launch the Training Job

   Call POST /v1/training/jobs or use the Dashboard. You receive a job_id immediately. Billing is per-GPU-hour from the moment the job enters running.

4. Monitor Progress in Real Time

   Stream loss, accuracy, gradient norm, and GPU utilisation through the WebSocket endpoint or refresh the Dashboard metrics view. Set threshold-based Slack / email alerts on failure.

5. Export the Trained Model

   On succeeded status, call GET /v1/training/jobs/{id}/export to download the weights + config bundle. You can optionally register the exported model as a custom identifier on /v1/models and run inference against it.

Production API Integration Checklist

Before shipping to production, verify every item below:

• ✅ API key stored in a secrets manager (NOT hardcoded, NOT in client-side JS)
• ✅ Requests set timeout <= 120 s and retry on 429/502 with exponential backoff
• ✅ Responses handle all documented error codes
• ✅ Streaming endpoints consume SSE per the Server-Sent Events spec
• ✅ Large outputs use chunked transfer; keep per-request payload < 6 MB
• ✅ X-Idempotency-Key sent for every financial/billing-linked write call
• ✅ Dashboard alerts configured at 80% and 95% of annual Token quota
• ✅ Technical support contact saved in your on-call rotation

Support & Contract References

Compliance Evidence Package

Supporting records that accompany the product documentation and validate end-to-end operational delivery. Sections marked as placeholders should be filled with actual company records before audit or compliance review.

6. Service Delivery Records

Every customer engagement generates five mandatory delivery artefacts — listed below with their recommended retention policy.

• Customer Account Creation[Insert Actual Record · PDF / CRM Export]
• API Credential Issued[Insert Actual Record · Key Log + Date]
• Service Activation Date[Insert Actual Date · Contract Clause Ref]
• Usage Records[Insert Actual Logs · CDR / API Log Export]
• Technical Support Records[Insert Ticket Logs · Zendesk / Jira Export]

7. Recommended Supporting Documents

The following documents, when bundled together, form the complete compliance evidence package for PREBO TRADING LIMITED's AI Computing Platform service.

8. Business Classification Statement

PREBO TRADING LIMITED provides technology services across the following regulated business categories. Each category maps to supporting evidence artefacts listed above.

• Artificial Intelligence Computing Services — Inference + Training API service delivery records, usage logs, SLA reports
• Cloud Infrastructure Services — Cloud invoices, server asset list, deployment region records
• Software Platform Services — Product documentation, API reference, release notes
• Data Processing Services — Training ETL logs, DPA, data retention & deletion records
• Technology Integration Services — Support tickets, onboarding notes, architecture review artefacts