​
Confidential Generative AI APIs

Atoma’s on-demand offerings provide immediate access to state-of-the-art generative AI models through privacy-preserving APIs that guarantee confidentiality for both system prompts and user data. Our platform leverages composable encryption technology to enable unprecedented levels of privacy protection, allowing service providers and end-users to maintain control over their respective sensitive information.

​
Core Capabilities

Multi-Modal AI Processing
  • Text generation and completion (e.g. gpt-oss, Llama, Mistral, DeepSeek, Qwen, Kimi models).
  • Image generation and editing (Flux, Stable Diffusion, etc).
  • Audio processing and speech synthesis.
  • Video analysis and generation.
  • Code generation and completion.
Retrieval-Augmented Generation (RAG)
  • Private document embedding and indexing.
  • Confidential knowledge base querying.
  • Secure vector database operations.
  • Private semantic search capabilities.
High-Performance Inference
  • Sub-second response times for most queries.
  • Batch processing for large-scale operations.
  • Real-time streaming responses.
  • Auto-scaling based on demand.

​
Composable Encryption: Dual Privacy Protection

Atoma’s breakthrough composable encryption technology enables simultaneous protection of both system-level prompts (service provider IP) and user prompts (customer data), addressing a critical gap in current AI privacy solutions.

​
How It Works

​
Dual-Layer Privacy Architecture

Layer 1: System Prompt Protection
  • Service providers’ proprietary prompts, instructions, and system configurations remain encrypted
  • AI service companies can protect their prompt engineering IP and competitive advantages
  • System-level context and fine-tuning parameters are never exposed
Layer 2: User Data Protection
  • End-user prompts, documents, and responses are encrypted with user-controlled keys
  • Customer data remains confidential from both Atoma and the service provider
  • The combination of advanced cryptographic primitives and trusted execution ensure only the user can decrypt their data

​
Real-World Use Case: Cursor IDE

Consider how Cursor IDE could leverage Atoma’s composable encryption: Cursor’s Protected Assets:
  • Proprietary system prompts that make Cursor’s AI unique
  • Custom code analysis algorithms and instructions
  • Specialized programming assistance methodologies
  • Training data and model fine-tuning parameters
Developer’s Protected Assets:
  • Private repository code and intellectual property
  • Confidential business logic and algorithms
  • Sensitive customer data within codebases
  • Personal coding patterns and preferences
Result: Cursor can offer AI-powered development assistance while protecting both their competitive AI technology and their customers’ proprietary code.

​
Enterprise-Grade API Features

​
OpenAI-Compatible Interface

# Standard OpenAI API call with Atoma's confidential computing
curl https://api.atoma.network/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $ATOMA_API_KEY" \
  -d '{
    "model": "gpt-oss-120b",
    "messages": [
      {"role": "system", "content": "Your protected system prompt"},
      {"role": "user", "content": "User query with private data"}
    ],
    "confidential": true,
    "user_encryption_key": "user_public_key"
  }'

​
Advanced Privacy Controls

Encryption Key Management
  • User-controlled encryption keys for data sovereignty.
  • Hardware Security Module (HSM) key storage.
  • Automatic key rotation and versioning.
  • Multi-party key escrow options.
Audit and Compliance
  • Comprehensive audit logs with cryptographic integrity.
  • Close to real-time attestation of secure execution.
  • Compliance reporting for GDPR, HIPAA, SOX.
Access Controls
  • Fine-grained permission management.
  • Role-based access to different encryption layers.
  • Temporary access tokens with automatic expiration.
  • Multi-factor authentication for sensitive operations.