Architecture

Prism is a comprehensive multi-party computation platform that provides a computation management service for confidential computing using Trusted Execution Environments (TEEs). It integrates with CocosAI to enable secure, private computations across multiple parties while maintaining data confidentiality and computation integrity.

Overall System Architecture

Prism operates as a cloud-native platform that orchestrates confidential computations across distributed TEE-enabled infrastructure. The system is designed with a modular, microservices architecture that provides scalability, security, and multi-tenancy support.

Key Architectural Principles

• Zero-Trust Security: All communications are encrypted and authenticated using mTLS
• Multi-Tenancy: Workspaces provide isolated environments for different organizations
• TEE-Agnostic: Supports multiple TEE technologies (AMD SEV-SNP, Intel TDX) through the CocosAI backend
• Attestation-First: All computations require cryptographic proof of TEE integrity
• Bring Your Own CVM: Flexible infrastructure deployment options

System Components

Prism Cloud Services

Service	Description	Key Responsibilities
Auth	Authentication and authorization management	- Policy enforcement for backends, certs, and computations - Role-based access control - API token management
Users	User identity and workspace membership	- User registration and authentication - Workspace association and roles - Integration with SuperMQ users service
Computations	Computation lifecycle management	- Computation definition and metadata - Participant coordination - Execution orchestration and monitoring
Backends	TEE infrastructure management	- CVM provisioning and lifecycle - Connection to CocosAI agents - Multi-cloud and on-premise support
Certs	Certificate Authority services	- mTLS certificate issuance and management - Certificate revocation and renewal - OCSP endpoint for certificate validation
Billing	Resource usage and payment management	- Usage tracking and billing - Resource access control based on plans - Multi-tenant cost allocation
Workspaces	Tenant isolation and resource organization	- Domain separation and access control - Resource grouping and management
User Interface	Web-based platform access	- Visual interface for all platform operations

CocosAI TEE Runtime Components

Prism integrates with CocosAI components that run on TEE-enabled hardware:

Component	Location	Description
Manager	TEE Host	- CVM provisioning and lifecycle management - IGVM/OVMF firmware configuration - TEE monitoring and secure destruction
Agent	Inside CVM/TEE	- Computation execution coordination - Secure data and algorithm ingestion - Remote attestation report generation - Secure result extraction
EOS	Inside CVM/TEE	- Lightweight Linux distribution (Buildroot-based) - Minimal attack surface - Container runtime for workloads
CLI	Client/Admin	- Administrative interface to TEE operations - Attestation verification - Algorithm and data provisioning

Data Flow and Interaction Patterns

Computation Lifecycle

1. Computation Definition

   • User defines computation via Prism UI or API
   • Specifies participants, data providers, algorithm providers
   • Sets privacy policies and result distribution rules
   • Participants provide public keys to be used for identification by agent

2. Backend Provisioning

   • Prism Backends service selects appropriate TEE infrastructure
   • Requests CVM creation from CocosAI Manager or public cloud CVM
   • Establishes secure communication channels

3. TEE Initialization

   • Prism Certs service issues mTLS certificates for secure communication between agent and prism.
   • Manager provisions CVM with IGVM/OVMF firmware. Alternatively CVM is provisioned on the public cloud
   • EOS boots inside TEE with Agent
   • Agent establishes secure channel with Prism

4. Participant Onboarding

   • Agent starts and prepares to receive assets
   • Attestation reports verified before participation

5. Secure Data Ingestion

   • Algorithm providers submit algorithms
   • Data providers upload datasets
   • All transfers use end-to-end encryption (using mutual attested TLS)

6. Computation Execution

   • Agent coordinates execution inside TEE
   • Workloads run within EOS
   • Progress monitored by Prism Computations service

7. Result Distribution

   • Results extracted from TEE
   • Distributed to authorized recipients only
   • Audit trail maintained for compliance through computation logs and events

8. Secure Cleanup

   • TEE securely destroyed after computation
   • All intermediate data wiped
   • Certificates revoked and resources deallocated

Communication Protocols

• Prism ↔ CocosAI Manager: gRPC over mTLS
• Agent ↔ Prism: gRPC over mTLS
• Agent ↔ CLI: gRPC over mTLS with attestation
• Client ↔ Prism: HTTPS with OAuth2/JWT
• Prism Internal: Service mesh with mTLS

Security Architecture

Trust Model

• Hardware Root of Trust: TEE hardware provides foundational security
• Attestation Chain: Cryptographic proof of TEE integrity from hardware to application
• Zero-Knowledge Coordination: Prism coordinates computations without accessing data
• Certificate-Based Identity: All entities authenticated via PKI

Attestation Flow

1. TEE Measurement: Hardware generates cryptographic measurements of CVM state
2. Report Generation: Agent produces signed attestation report
3. Verification: CLI validates report against expected measurements
4. Policy Enforcement: Only verified TEEs allowed to participate

Data Protection

• Data in Transit: End-to-end encryption using mTLS
• Data at Rest: Encrypted storage with key management
• Data in Use: Protected by TEE hardware encryption
• Key Management: Distributed key derivation without central key storage

Deployment Architectures

Cloud Deployment

• Prism services deployed on major cloud providers (GCP, Azure)
• CocosAI backends on TEE-enabled cloud instances
• Managed certificates and load balancing

On-Premise Deployment

• Self-hosted Prism services in customer infrastructure
• Direct control over TEE hardware
• Air-gapped environments supported

Hybrid Deployment

• Prism control plane in cloud
• TEE compute resources on-premise
• Secure tunneling for coordination

Scalability and Performance

Horizontal Scaling

• Microservices architecture enables independent scaling
• Load balancing across multiple TEE backends
• Database sharding for multi-tenant workloads

Limitations

• TEE Resource Constraints: Memory and CPU limitations within enclaves
• I/O Overhead: Encryption/decryption costs for large datasets
• Hardware Dependencies: Requires specific TEE-capable processors
• Network Latency: Multi-party coordination introduces communication delays

Monitoring and Observability

System Metrics

• CVM health and resource utilization
• Computation execution progress and performance
• Certificate lifecycle and validity status
• Backend connectivity and availability

Security Monitoring

• Attestation verification
• Certificate validation events
• Unauthorized access attempts
• Anomalous computation patterns

Audit Trail

• Complete computation lifecycle logging
• Participant actions and data access
• Security events and policy violations
• Compliance reporting capabilities

Integration Points

APIs and SDKs

• REST APIs: Comprehensive API coverage for all platform operations
• CLI Tools: Command-line interface for automation and scripting

This architecture enables Prism to provide a comprehensive, secure, and scalable platform for confidential multi-party computation while maintaining the highest standards of data privacy and security through TEE technology.