Confidential AI Cloud · Phala

Description

Phala Cloud offers a confidential AI platform designed for running AI workloads with enhanced privacy and security. It leverages Trusted Execution Environments (TEEs) on GPUs, such as Intel TDX and NVIDIA H100/H200, to ensure that sensitive data and AI models remain private during processing. This allows for private LLM inference, sealed agents, and verifiable compute.

The platform supports the deployment of existing Docker Compose workloads, making it easy for developers to migrate their applications. Phala provides runtime measurements and cryptographic attestations, enabling users to verify the integrity and security of their AI workloads. This verifiable execution is a key feature, allowing users to trust the results of their computations.

Phala Cloud offers a GPU marketplace where users can launch H100, H200, and B300 GPU capacity with TEE-backed runtime proof and public attestations. The platform supports various AI models, including OpenAI-compatible LLM endpoints with private prompts and verifiable runtime state. Phala also provides an agent sandbox for running agent backends in a confidential VM with sealed keys, private memory, and verifiable execution.

The platform is designed for enterprise-grade compliance and security, with SOC 2 Type I certification, HIPAA compliance, and ISO 27001 certification in progress. Phala has a 99.9% uptime SLA and is GDPR compliant. The platform is trusted by over 5,000 users and industry leaders worldwide. Phala's confidential AI has been used in financial services, healthcare research, and AI SaaS platforms, demonstrating its versatility and reliability.

Confidential AI Cloud's Core Features

• Confidential AI execution on TEE-protected GPUs

• Private LLM inference

• Sealed agents

• Verifiable compute

• Support for Intel TDX and NVIDIA H100/H200 GPUs

• Deployment of existing Docker Compose workloads

• Runtime measurements and cryptographic attestations for verification

• GPU marketplace for launching GPU capacity

• OpenAI-compatible LLM endpoints

• Agent sandbox for running agent backends

• SOC 2 Type I certified and HIPAA compliant

• 99.9% uptime SLA

• GDPR compliant processing

• Enterprise-grade security and compliance

How to use Confidential AI Cloud?

1. 1

   Install: Install the Phala CLI using npm.

2. 2

   Deploy: Deploy your Docker workload using the Phala CLI.

3. 3

   Configure: Configure your application within the Phala environment.

4. 4

   Verify: Inspect the status, logs, and attestation from the command line.

5. 5

   Use: Access and utilize your deployed confidential AI application.

6. 6

   Monitor: Monitor the application's performance and security.

Confidential AI Cloud's Use Cases

• Private LLM Inference
• Sealed Agents
• Verifiable Compute
• Financial Services
• Healthcare Research
• AI SaaS Platform

FAQ from Confidential AI Cloud

• What is a Trusted Execution Environment (TEE)?

  A TEE is a secure area inside a processor that protects code and data from the operating system, hypervisor, and other applications.

• How does confidential AI protect sensitive data?

  Sensitive data and AI models remain private during processing by running inside hardware-backed secure environments.

• Is Phala compatible with existing AI frameworks?

  Yes, Phala supports existing Docker services and popular AI frameworks including TensorFlow, PyTorch, and Hugging Face.

• What are the performance implications?

  Confidential GPU workloads typically target near-native performance, with roughly 5-10% overhead depending on workload and hardware.

• How can I verify the security of my AI workloads?

  Phala exposes cryptographic attestations so users and systems can verify the workload and runtime state.

• How do I get started?

  Install the Phala CLI, deploy a Docker workload, then inspect status, logs, and attestation from the command line.

• What kind of models are supported?

  Phala supports various models including Qwen, Gemma, DeepSeek, MoonshotAI, and Z.ai.

• What is the pricing model?

  Pricing varies based on GPU type and usage, starting from $3.20/GPU/hr for H200 US.

• What compliance standards does Phala adhere to?

  Phala is SOC 2 Type I certified, HIPAA compliant, and has ISO 27001 certification in progress.

• What is the uptime SLA?

  Phala offers a 99.9% uptime SLA.