New

Build with encrypted inference

Explore Enclave API

Frontier
Sovereign AI

Prem delivers sovereign AI that is verifiable, private, and compounding.

An aerial view of a lush green valley

Where shall we begin?

A woman working on a laptop at a table beside a window

Designed & powered in Switzerland

SOC 2 Type I, ISO 27001 and FADP

Prem delivers sovereign AI that is verifiable, private, and compounding.

Run frontier models on encrypted GPUs with hardware-verified attestation at every layer.

Even Prem can’t access your data.

  • Private

    Deploy private agents. Your files, apps and instructions, in one private workspace.

  • Verifiable

    Access encrypted GPUs. Verify your data's privacy instantly with cryptographic attestation.

  • Compounding

    Build compounding intelligence. It learns how you work and gets better every week.

Sovereign intelligence at scale

<40ms
Latency
Dedicated GPU clusters, optimized for <40ms inference.
Zero
Data retention
The enclave wipes all plaintext from memory.
5+
Model families
Query the world’s leading open models through a unified endpoint.
99.982%
Uptime SLA
Proudly hosted in Switzerland, the world’s best address for data.
Evening light on the Alpstein range, Switzerland

Fluso

Deploy private agents.

Agent platform to build and deploy enterprise agents based on proprietary SOPs.

Where shall we begin?

How it works

From a sentence to a teammate.

Each agent is a conversation. Set it up once, then chat with it, or see what it did last.

  1. 01

    Connect your tools

    Gmail, Drive, Slack, Notion, Linear. An agent only sees what you connect.

  2. 02

    No-code agent builder

    Tell the manager what you need. It drafts the agent: instructions, apps, skills.

  3. 03

    Proactive event/schedule triggers

    Add a routine and the agent runs on its own. Hourly, daily, weekly, or whenever you ask.

Private by design. Transparent by default.

  • Open-source intelligence

    Built on open-source models. Inspect every layer. Your data never enters training.

  • Deployable boundaries

    Our secure cloud, your VPC, or fully air-gapped on premise. Same workspace, your rules.

  • EU data protection

    Hosted on Prem AI's confidential compute in the EU. GDPR-bound and sovereign by design.

Jobs worth handing off.

A few agents people set up in their first week.

  • Reads overnight mail and today's calendar, posts a short brief to Slack.

    Daily, 08:00
    • Gmail
    • Calendar
    • Slack
  • Looks for flights on a schedule and pings you when a fare drops.

    Hourly
    • Web
  • Searches the web for competitor news and updates the Notion page, with sources.

    Mondays, 09:00
    • Web
    • Notion
  • Pulls the numbers from Sheets and ships the update PDF to investors.

    Fridays, 16:00
    • Sheets
    • Drive
0 / 0
Meet your agents

Morning brief

Daily, 08:00

Running

3 replies need you, 2 meetings moved, the offsite budget is approved.

  • Gmail
  • Calendar
  • Slack
An aerial view of a lush green valley

EnclaveFor Developers

Build with encrypted inference

Run frontier models on encrypted GPUs with hardware-verified attestation at every layer. Even Prem can't access your data.

Developers

Fully OpenAI-compatible. Just swap your client SDK and go live.

Requests

XChaCha20-Poly1305

The TypeScript SDK encrypts the request payload before network egress.

import { createRvencClient } from "@premai/api-sdk"; const client = await createRvencClient({  apiKey: process.env.PREM_API_KEY,  clientKEK: process.env.CLIENT_KEK, // You generate this. We never see it.}); const response = await client.chat.completions.create({  model: "deepseek-v4.1-flash-1017",  messages: [{ role: "user", content: "Hello, privately." }],}); console.log(response.choices[0]?.message?.content);

DeepSeek V4.1 Flash · 1017

deepseek-v4.1-flash-1017
  1. Trusted Execution Environments generate a cryptographic proof of their identity and integrity. Your device verifies this proof before sending any data.

    ChallengeCPU enclaveGPU enclaveSession
Encrypted payload
Reply

Illustration of the request path. This demo sends no request.

Standard AI APIs require a leap of faith. We seal it shut.

Run AI models within hardware-enforced Trusted Execution Environments (TEEs). Replace legal promises with safeguards rooted in silicon.

Attestation

Verifying
  • ChallengeRandom nonce
  • CPU enclaveAMD SEV-SNP or Intel TDX report
  • GPU enclaveNVIDIA token signature and measurement
  • Sessionx-session-id, single use, five-minute TTL

Verified on your device by @premai/reticle: Rust, compiled to WebAssembly.

How attestation works

Attestation removes the need to take our word for anything.

Every deployed PCC enclave generates a cryptographic attestation report. This ensures genuine hardware verification and tamper-proof code execution before any data transfer is initiated.

DeepSeek V4.1 Flash · 1017

Hello, privately.

Attesting
  1. Attestation
  2. Client-side encryptionXChaCha20-Poly1305
  3. Gateway
  4. Secure executionCPU + GPU
  5. ResponseReply

Encrypted payload

Every payload processed in isolated enclaves, with keys only you control.

Payloads are encrypted locally using a public key bound to a verified enclave. Execution state and memory pages never leave the PCC enclaves unencrypted.

Through the local Confidential Proxy, with an OpenAI client.

import OpenAI from "openai"; const client = new OpenAI({  apiKey: process.env.PREM_API_KEY!,  baseURL: "http://127.0.0.1:8787/v1",}); const stream = await client.chat.completions.create({  model: "deepseek-v4.1-flash-1017",  messages: [{ role: "user", content: "Hello, privately." }],  stream: true,}); for await (const chunk of stream) {  process.stdout.write(chunk.choices[0]?.delta?.content || "");}

Fully OpenAI-compatible. Just swap your client SDK and go live.

Send chat and audio requests with client-side encryption through the Prem SDK or local proxy.

Case studies

Trusted by leading financial, healthcare and retail organizations

  • EU RegTech Leader

    The Prime Tower over the roofs of Zurich at sunset
    2,000+Regulatory documents per quarter
    Challenge
    EU data residency laws meant regulatory reviews couldn’t touch public cloud AI. Compliance teams processed 2,000+ regulatory documents per quarter entirely by hand, averaging 45 minutes per document.
    Solution
    Prem deployed a sovereign model inside the client’s perimeter. It now processes sensitive regulatory datasets hourly with zero data leaving the environment.
  • North American Healthcare Provider

    The old town of Basel across the Rhine in evening light
    10M+Patient records
    Challenge
    10M+ patient records locked inside unstructured PDFs. HIPAA rules made public cloud AI a non-starter, so critical insights stayed buried in manual workflows.
    Solution
    Prem’s sovereign infrastructure let clinical teams extract structured intelligence from raw documents, cutting processing time without adding security or compliance risk.
  • Fortune 50 Retailer

    A cobbled lane of shops in the old town of Zurich on a rainy day
    3,000+Stores, and 15 data systems
    Challenge
    Customer behavior data spread across 3,000+ stores and 15 data systems. The team needed AI-driven demand forecasting but couldn’t send proprietary sales patterns to a third-party API.
    Solution
    Prem deployed an on-premise forecasting model trained on 3 years of proprietary transaction data, running entirely within the retailer’s existing infrastructure.
0 / 0

Developer resources

Aerial photography of green mountains covered with green trees at daytime
bash

Install the SDK

npm install @premai/api-sdk

Get started

You can integrate Prem API into an application, or you can evaluate Prem API for your organization. The procedure is the same.

A lake between mountains on a quiet morning in the Swiss Alps

Designed & powered
in Switzerland

Proudly hosted in Switzerland, the world’s best address for data.

Request a demo
Lugano, Switzerland

Frequently asked questions

  • PCC offers an OpenAI-compatible inference API. A key feature is the end-to-end encryption of your data, which is then processed securely within hardware-sealed Trusted Execution Environments (TEEs).

  • No. The enclave images contain no SSH access, no debug ports, and no administrative backdoors. This is not a policy we follow. It is a constraint the hardware imposes.

  • We do not use your data for training, fine-tuning, or plaintext logging. Your information only exists as plaintext within the secure enclave for the duration of your request.

  • Data is encrypted on your device using XChaCha20-Poly1305 before transmission. Plaintext data is securely contained in only two locations: your device and the sealed hardware environment.

  • Attestation removes the need to take our word for anything. It is hardware-signed proof of exactly what code is running inside the enclave.

  • Yes. The SDK performs attestation verification by default before every request. The verification stack is open-source.

  • No, the impact is negligible. The CVM imposes a negligible compute overhead, adding approximately 5% on current hardware.

  • PCC's architecture streamlines compliance with major global privacy and security frameworks like the EU AI Act, GDPR, and ISO 27001. Its core value is eliminating the primary compliance challenge: exposing plaintext data to the AI provider.

Frontier Sovereign AI

Prem delivers sovereign AI that is verifiable, private, and compounding.

Even Prem can’t access your data.