Documentation

Technical documentation for the Voban platform. Some features are currently being deployed.

Overview

Voban gives companies controlled access to AI models. No proprietary model. A control framework: model selection, security rules, traceability, and data governance.

Core principles

• Control — Define your own AI usage rules
• Transparency — Every decision is traced and explainable
• Modular sovereignty — Adjustable per use case
• No lock-in — OpenAI-compatible API, interchangeable models

Architecture

Voban sits between users and AI models. Every request passes through the Guard before reaching the model.

User (Chat / API / IDE)
       |
   [ Voban Guard ]
   - Analyze request
   - Enforce rules
   - Select model
       |
 Model (sovereign or frontier)
       |
   [ Voban Guard ]
   - Filter response
   - Log event
       |
   User

Components

• Gateway — Single entry point, OpenAI-compatible
• Guard — Security and policy layer
• Router — Directs requests based on rules
• Model Connectors — Open source and frontier providers
• Admin API — Configuration and monitoring

Models

The admin activates available models per workspace.

Sovereign models run on French infrastructure. Pragmatic models are hosted in the EU by their providers (US entities).

Sovereignty

Voban offers a sovereignty dial adjustable per use case.

Sovereign mode

• Open source models (Mistral, Qwen, Llama)
• French infrastructure
• No non-EU dependency for data processing
• Recommended: HR, strategy, personal data, legal

Pragmatic mode

• Frontier models (Claude, GPT)
• EU-hosted by providers
• GDPR via contractual agreements (DPA), not jurisdiction
• Recommended: content, code, translation, summaries

Guard

Security and policy layer on every request. Transparent to users.

How it works

1. Inbound — Analyze, detect PII, apply routing, mask if needed
2. Outbound — Check response, log full event

API

OpenAI-compatible API. Any compatible client works without modification.

Authentication

Authorization: Bearer voban_sk_...

Chat Completions

POST /v1/chat/completions

{
  "model": "mistral-large",
  "messages": [
    { "role": "user", "content": "Summarize this document." }
  ]
}

List models

GET /v1/models

{
  "data": [
    { "id": "mistral-large", "mode": "sovereign" },
    { "id": "llama-3.3", "mode": "sovereign" },
    { "id": "claude-sonnet", "mode": "pragmatic" }
  ]
}

IDE integration

OPENAI_API_BASE=https://api.voban.ai/v1
OPENAI_API_KEY=voban_sk_...

Works with Cursor, OpenCode, Continue, and any OpenAI-compatible tool.

Chat

Built for daily team use within a validated framework.

• Multi-turn conversations with history
• File uploads (PDF, images, text)
• Admin-configured model selection
• Guard runs transparently in the background

Admin

Control AI deployment across your organization.

Models

• Enable/disable models per workspace
• Set default model
• Configure sovereignty per use case

Guard rules

• PII policies per user profile
• Sovereign routing config
• Injection detection toggle

Usage

• By collaborator, team, model
• Budgets per user type
• Guard event log

Security & compliance

• Platform operated in Europe
• Sovereign models on French infrastructure
• Data not used for training
• TLS 1.3 in transit, AES-256 at rest
• GDPR — EU processing, DPA available
• AI Act — compliance support planned
• NIS2 — built-in traceability

Roadmap

Questions? contact@voban.ai