Sovereign AI governance platform
Unlock the potential of Generative AI.
Keep absolute control.
The sovereign platform that lets CIOs and CISOs accelerate AI adoption without compromise. Give your teams the best AI tools on the market while blocking data leaks, keeping your budget under control (FinOps) and guaranteeing your compliance.
Barriers removed
Turn AI risk into a competitive advantage
The question is no longer whether your employees use AI, but how to give them a trusted framework to innovate safely.
Become the AI provider
Instead of enduring Shadow AI (ChatGPT, Claude), offer a powerful internal alternative. IT regains control over the flows while meeting business needs.
Sovereignty by design
Protect your trade secrets. Names, IBANs and confidential data are pseudonymized on the fly before they ever leave your infrastructure.
Predictability and ROI (FinOps)
No more budget overruns driven by token consumption. Allocate budgets per team, cap costs and rationalize cloud API usage.
Our solutions
Two complementary building blocks, total control
Deployed on-premise or on your sovereign cloud. One command to install, all AI traffic funneled through your gateway, and an IT department that gets its peace of mind back.
Spinlayer-chat
The enterprise knowledge IDE. A sovereign assistant that queries your internal document bases, cites its sources, and lets no data leave your perimeter.
- Hybrid RAG architecture contained on your servers
- SSO / RBAC integration (Keycloak, Active Directory)
- Multi-source ingestion (NAS, SharePoint, sovereign S3)
- Mandatory source citation with the exact paragraph
- Fine-grained system-prompt tuning per organization
Spinlayer-gateway
The AI security and financial-governance gateway. It sits between your applications and AI providers to filter, anonymize and rationalize every request.
- Stateless in-memory PII pseudonymization (< 12ms)
- Anti-injection firewall (98.2% jailbreak detection)
- FinOps dashboard: token quotas per user / API key
- Universal API compatible with the OpenAI format
- Automatic fallback to local open-source models
Architecture
A sovereign intermediation layer at the edge of your IT system
Spinlayer deploys as a reverse proxy between your applications and AI providers. No flow leaves your perimeter without being analyzed, filtered and logged.
Every AI request must transit through the sovereign gateway. The Guard analyzes and filters in memory (stateless) before anything is transmitted to the models — local or cloud.
Use cases
Concrete, realistic, pragmatic deployments
No marketing promises. Here are typical Spinlayer deployment scenarios solving real compliance, control and sovereignty challenges.
Securing access to external LLMs
Finance · Insurance · Financial servicesContext: Analysts, advisors and developers frequently use public generative AI to summarize reports, analyze data or generate code.
Problem: Critical risk of massive data leaks. Without strict network control, employees paste sensitive information into prompts — account numbers, tax data, API keys, trade secrets.
Spinlayer solution: All outbound flows centralized through Spinlayer-gateway. Every prompt passes through the Guard engine, which detects and replaces personal data with anonymous tokens ([REDACTED_IBAN], [REDACTED_NAME]) before transmission. On the way back, the gateway reinjects the original data. No identifying data leaves the perimeter, Guard latency < 12ms.
A knowledge IDE for HR and Legal departments
Large companies · Ministries · Local authoritiesContext: HR and Legal teams spend hours searching for answers across thousands of pages of regulations, employment contracts and collective agreements.
Problem: Cloud AI is off-limits for professional-secrecy and sovereignty reasons. Classic RAG solutions often force you to host documents on third-party servers.
Spinlayer solution: On-premise deployment of Spinlayer-chat. Documents are indexed locally. The assistant always cites the source document and the exact paragraph. 100% of the data stays contained. Document research time divided by 4.
Protection against prompt-injection attacks
Software vendors · DevOps teamsContext: Your organization exposes an AI agent (support chatbot, automation tool) to customers or external users.
Problem: Prompt-injection attacks. Malicious users manipulate the chatbot with hidden instructions to extract system data or bypass security rules.
Spinlayer solution: Spinlayer-guard in strict cut-off between the application and the model. Its specialized ONNX model analyzes prompt intent. If an injection is detected, the request is blocked at the boundary without ever reaching the LLM. 98.2% of known techniques blocked, token savings on malicious requests.
Optimizing technical support and incident management
Helpdesks · MSSPs · Operations centersContext: First-level technicians spend time digging through resolved-ticket history, writing reports and qualifying incidents for escalation.
Problem: Classic keyword search is slow and ineffective against the diversity of phrasings. Processing times stretch, wrong escalations multiply, and technical knowledge is lost to turnover.
Spinlayer solution: Spinlayer-chat connected to resolved-incident history. The technician describes the problem in natural language and gets the probable cause with sourced remediation steps. Substantially reduced MTTA, automated knowledge capitalization into knowledge-base articles.
Ecosystem
Deploy Spinlayer with an integration partner
Spinlayer relies on a network of specialized consulting and integration partners to help you deploy a sovereign AI environment tailored to your technical and regulatory context.
Specialist in sovereign cloud engineering, DevOps and infrastructure security. Opsky orchestrates automated on-premise installation, operational maintenance (MCO) and the hardening of Spinlayer-gateway.
Specialist in cybersecurity, networks and the digital workplace. Anetys guarantees airtight, compliant integration of Spinlayer into enterprise architectures with high security requirements.
Are you a consulting firm or integrator interested in joining the Spinlayer ecosystem?
Become a partnerFrequently asked questions
No magic jargon, factual answers
Where does my data transit and where is it stored with Spinlayer?
Your data stays 100% inside your network perimeter. Spinlayer is designed to be deployed either on-premise or on your sovereign private cloud (Scaleway, OVHcloud, Outscale, AWS, Azure, GCP). No data flow, no prompt and no indexed document ever transits through Spinlayer's servers or non-European infrastructure, unless you explicitly configure a gateway to a third-party LLM.
Does Spinlayer use our data to retrain models?
No. This is a founding principle of our architecture. Spinlayer never records, stores or transmits any prompt or company data for training purposes. The local models used for anonymization process flows in memory (stateless) and release them immediately after processing.
How does Spinlayer support GDPR and EU AI Act compliance?
Spinlayer acts as an automated compliance barrier at the edge of your information system. For GDPR, Spinlayer-guard intercepts and anonymizes personally identifiable information (PII) before any external processing. For the EU AI Act, the solution natively provides mandatory traceability, bias-risk management, malicious-injection blocking and full flow logging.
What is the latency impact on requests?
The impact is transparent for the end user. The Guard engine is optimized in ONNX format to run locally on CPU or a light GPU. Analyzing, detecting and anonymizing a standard prompt takes less than 12 milliseconds. Overall latency then depends exclusively on the chosen LLM.
How does data-leak and injection detection work?
Spinlayer-guard does not rely on simple regular expressions, which are easy to bypass. It combines lightweight local named-entity recognition (NER) models and text-structure classifiers trained specifically to detect manipulation attempts. When a threat is detected, the flow is blocked or sanitized instantly at the network boundary.
What kinds of sensitive data can be filtered?
The system recognizes more than 40 categories: credit card numbers, IBANs, social security numbers, credentials (API keys, passwords), health data, proper names and addresses. You can also configure your own business dictionaries (secret project names, proprietary source code).
Which AI models and providers are compatible?
The gateway is universal. It exposes a standardized API compatible with the OpenAI format. It can route flows to cloud providers (OpenAI, Anthropic, Mistral AI, Azure OpenAI) or to open-source models hosted locally (via Llama.cpp, vLLM, Hugging Face, or your private GPU servers).
Does Spinlayer-chat embed local AI models?
No. Spinlayer-chat is exclusively a sovereign frontend client (a "knowledge IDE"). It does not manage any model itself. It connects to Spinlayer-gateway, which queries your AI providers. This split keeps the interface light and guarantees security tightness through Spinlayer-guard's systematic filtering.
Ready to take back control of AI in your organization?
Get in touch with our engineers to plan the integration of Spinlayer into your infrastructure.
Or contact us directly by email: contact@spinlayer.fr