An introduction to the Prem product stack
DSP · May 2026 · Confidential
About Prem
Prem builds sovereign AI infrastructure that keeps your data private — end-to-end encrypted, hardware-enforced, and post-quantum ready.
From trusted hardware to agentic AI applications, every layer of the Prem stack is designed so that you own every bit of your intelligence.
Switzerland · USA · Italy
The Prem Stack
Each layer depends on the one below. The customer owns every layer.
Layer 3 — Application
Prem Fluso + VEREX
Agentic AI workspace. Verifiable reasoning. Self-improving agents, institutional memory, autonomous fine-tuning.
Layer 2.5 — Voice
Prem Sotto
Sovereign voice-to-text. <200ms latency. Zero retention. E2E encrypted. Self-hostable.
Layer 2 — Developer Interface
Prem API
OpenAI-compatible API with E2E encryption, hardware attestation, post-quantum cryptography. Drop-in replacement.
Layer 1 — Foundation
Prem Compute
Hardware-attested AI infrastructure. AMD SEV-SNP, Intel TDX, NVIDIA Confidential Computing. On-prem, air-gapped, or hybrid.
Sovereign intelligence requires a full stack built in a specific order.
Trusted hardware at the base. Private inference in the middle. An application layer that compounds intelligence on top.
The compounding engine runs for the customer. Every workflow, decision, and correction feeds a proprietary intelligence asset owned entirely by the deploying organisation.
Layer 1 — Foundation
Hardware-attested AI infrastructure you can trust.
AMD SEV-SNP, Intel TDX, NVIDIA Confidential Computing — processing inside tamper-proof enclaves.
On-premise, air-gapped, or hybrid. Runs on your infrastructure, in your data center, under your control.
Cryptographic proof that your data runs where you expect — verifiable at every layer.
Layer 2 — Developer Interface
OpenAI-compatible. End-to-end encrypted. Drop-in replacement.
| Standard AI API | Prem API | |
|---|---|---|
| Who can see your data? | Provider, staff, cloud host | Only you |
| Server compromised? | Your data is exposed | Attacker sees encrypted bytes |
| How is privacy verified? | Trust the privacy policy | Hardware-signed cryptographic proof |
| Quantum-resistant? | Vulnerable in the future | Protected today |
| API compatibility | OpenAI-compatible | Same interface — no rewrite |
Layer 2.5 — Voice
Sovereign voice-to-text. Powered by Deepgram Nova-3.
<200ms
Latency
Zero
Data retention
E2E
Encryption
Self-hostable
Deployment
Layer 3 — Application
Agentic AI workspace with verifiable reasoning.
Autonomous agents that execute complex workflows — research, analysis, document generation — with full traceability.
Verifiable, explainable reasoning engine. Every decision produces an auditable evidence chain — no black boxes.
Every interaction, workflow, and correction compounds into a proprietary knowledge asset owned by your organization.
Self-improving models that learn from your corrections and adapt to your domain over time.
Why Sovereign
Your operational data, trade secrets, and strategic plans never leave your control. Processing happens in hardware-sealed environments.
GDPR, nFADP, SOC 2, and emerging AI regulations demand verifiable data protection — not promises.
Every AI interaction builds your proprietary intelligence. This becomes a defensible competitive moat over time.
Deployment
Run the full stack inside your data center. No external network dependency. Full physical control.
Operate completely disconnected from the internet. Ideal for defense, critical infrastructure, and high-security facilities.
Sensitive workloads on-prem. Less critical tasks in the cloud. Orchestrated as one stack.
Quantum-resistant cryptography at every layer. Protected against harvest-now-decrypt-later threats.
Opportunity
Starting points for our conversation — not a prescription.
AI agents that analyze TOS data, optimize yard planning, and predict bottlenecks — with verifiable reasoning.
Capture decades of operator expertise into an institutional memory that trains the next generation.
Private AI that powers digital twin platforms — scenario planning without exposing operational data.
Self-improving agents that draft reports, audit trails, and regulatory submissions.
Real-time, multilingual voice-to-text for crane operators, planners, and ground crews.
AI agents trained on your TOS documentation — instant answers for your global helpdesk.
What are your biggest operational challenges today?
Where does data privacy matter most in your workflows?
What would you want an AI agent to handle for your team?
Marco Guerriero — Chief Growth & Operations Officer, Prem AI
marco.guerriero@premai.io
Emanuele Carpanzano — Chief Science Officer, Prem AI
emanuele@premai.io
premai.io · docs.prem.io · sotto.prem.io