Zum Hauptinhalt springen

FAQ zu dem LLMaaS-Cloud-Temple-Serviceangebot

Angebot und Fähigkeiten​

What is the value proposition of your LLMaaS service?​

Our service provides you with sovereign and secure access to a broad catalog of cutting-edge AI models via an OpenAI-compatible API. We handle all the complexity of the certified infrastructure SecNumCloud 3.2, allowing you to focus on delivering value for your applications, while maintaining cost control through a predictable economic model and accelerating your development with open standards.

What types of models do you offer, and how does the catalog evolve?​

We offer a rich catalog of 45 open-source models, including the families Llama (Meta), Qwen, Deepseek, Mistral, Gemma (Google), Cogito, and Granite (IBM). Our models are specialized for various use cases:

  • Complex reasoning and conversational tasks
  • Code generation and analysis
  • Long-document analysis (up to 128k tokens with models such as Gemma 3 or Qwen2.5-VL)
  • Multimodal capabilities (image analysis with the Granite Vision, Gemma 3, Qwen2.5-VL, and Mistral Small series)
  • Advanced multilingual support (with models such as Llama 3.3 and the Qwen family)

This catalog is reviewed quarterly to incorporate the most performant models, with a transparent lifecycle policy (3-month notice prior to any deprecation) to ensure stability for your applications.

What are the main functional capabilities of your API?​

Our API, 100% compatible with the OpenAI standard, enables you to easily build advanced features such as:

  • Conversational chatbots/assistants.
  • Retrieval-Augmented Generation (RAG) systems using our embedding endpoint.
  • Image and visual document analysis applications without requiring prior OCR.
  • Audio transcription via a dedicated service.

Sicherheit und Souveränität​

How do you ensure data sovereignty and confidentiality?​

Data sovereignty is at the core of our offering. It is based on several key pillars:

  1. Hosting in France: Our infrastructure is exclusively located in France and operated by Cloud Temple, a company governed by French law. This protects us from extraterritorial laws such as the U.S. CLOUD Act.
  2. SecNumCloud 3.2 Certification: The highest security certification granted by ANSSI ensures this positioning.
  3. No Data Retention: We do not store your prompts or responses. Data is processed temporarily in memory during inference only.
  4. End-to-End Encryption: All communications with the API are encrypted using TLS 1.3.

How do you manage access, permissions (RBAC), and authentication (SSO, MFA)?​

Our service integrates with your security policy according to a shared responsibility model:

  • For your end users: Identity management, permissions (RBAC), and strong authentication (SSO, MFA) are the responsibility of your application.
  • For your administrators: Access to the Cloud Temple Console for service management (key management, monitoring) can be connected to your enterprise SSO (via OpenID Connect) to leverage your existing security policies, including MFA.
  • For your applications: You can create different API keys with distinct consumption quotas to segment permissions at the application level.

How is the platform secured?​

We implement a defense-in-depth approach:

  • Strong authentication via API key for every request.
  • TLS 1.3 encryption for all data flows.
  • Network protection using next-generation firewalls and intrusion detection/prevention systems (IDS/IPS).
  • Application-level guardrails to protect against LLM-specific threats, such as prompt injection and model exploitation attempts.
  • Regular penetration testing conducted by qualified auditors (PASSI) as part of our SecNumCloud 3.2 certification process.

Integration und Verwendung​

How does your API integrate?​

Our service is "API-first." Being OpenAI API-compatible, you can use all standard SDKs (Python, Node.js, etc.) and frameworks such as LangChain or LlamaIndex for rapid integration. We provide comprehensive technical documentation (OpenAPI) and code examples.

Can the user experience be customized?​

Yes, absolutely. Since our service is a "headless" API, you have full control over the interface and end-user experience, including integration of your own design system and display of compliance messages, which are your responsibility.

You have full control over the models your application calls. For use cases sensitive to IP, we highlight IBM's Granite models, which come with a contractual, unlimited indemnification against IP claims — a unique guarantee in the market.

Economic Model and Support​

What is your business model?​

Our model is based on actual usage (tokens for language, minutes for audio) to reflect real computational costs. To ensure predictability and budget control, we offer Service Tiers with capped budgets and custom contracts. You can monitor your usage in real time via the Temple Cloud Console for full transparency. Additionally, the number of users is always unlimited, regardless of the plan you choose.

What level of support and SLA do you offer?​

Our standard offering includes technical support in French and English, accessible via our customer portal, with a commitment to first response within one business day. The platform comes with a 99% availability guarantee. For the most critical needs, our optional Professional Services provide enhanced SLAs, including priority support with response times of up to 4 hours for critical incidents, as well as tailored strategic support (innovation workshops, ITIL governance).

How can I monitor the health status of the LLMaaS platform?​

We believe in complete transparency regarding the availability of our services. You can check in real time the operational status of each AI model on our public status page: https://llmaas.status.cloud-temple.app/.

This page provides detailed information on:

  • The overall status of the platform.
  • The availability of each model individually.
  • Performance metrics such as time to first byte (TTFB) and throughput (tokens/s).
  • Incident history for comprehensive tracking.