What is a TEE-Protected Gateway?
RedPill’s core innovation is running the entire AI gateway infrastructure inside Trusted Execution Environments (TEE). Unlike competitors who only protect specific models, every single request flowing through RedPill is hardware-protected, regardless of which AI provider you use.
How It Works
1. Request Enters TEE
When you make an API request to RedPill:- Request data is encrypted in transit (TLS)
- Once inside TEE, data is decrypted only within the secure enclave
- Even RedPill operators cannot access the data
2. Processing in TEE
Inside the TEE enclave, the gateway:- Authenticates your API key
- Validates the request
- Routes to the appropriate LLM provider (OpenAI, Anthropic, etc.)
- Forwards your request through secure channels
3. Response Processing
When the LLM responds:- Response enters the same TEE enclave
- Processed and formatted within secure environment
- Encrypted before leaving TEE
- Delivered securely to you
TEE Technology
What is a Trusted Execution Environment?
A TEE is a secure area inside a processor that provides:Memory Isolation
Protected memory that other processes cannot access
Code Integrity
Guarantees code executes exactly as intended
Data Confidentiality
Encrypted data processing invisible to host system
Remote Attestation
Cryptographic proof of genuine TEE execution
Hardware-Level Protection
TEE protection is enforced by CPU hardware, not software:| Feature | Software Security | TEE Security |
|---|---|---|
| Protection Level | Application layer | Hardware (CPU) layer |
| Access Control | OS-based | CPU-enforced |
| Memory Encryption | Optional | Always on |
| Attestation | Not available | Cryptographic proof |
| Trust Model | Trust OS & cloud | Trust only CPU |
TEE Technologies Used
RedPill leverages industry-standard TEE technologies:- Intel SGX - Secure enclaves on Intel processors
- AMD SEV - Secure Encrypted Virtualization
- ARM TrustZone - ARM processor security
- NVIDIA H100 TEE - GPU TEE for confidential AI models
Privacy Guarantees
What RedPill Cannot See
Even with full system access, RedPill operators cannot see: ✅ Your API requests (prompts, messages) ✅ Model responses ✅ Function calling parameters ✅ Embeddings input text ✅ Image data (for vision models) ✅ Any data processed in TEEWhat RedPill Can See
For operational purposes only:- Encrypted request metadata (timestamp, model selected)
- Aggregated usage statistics
- Rate limit counters
- Error logs (no sensitive data)
Cryptographic Attestation
Verify TEE Execution
RedPill provides cryptographic proof that your request executed in genuine TEE:- TEE hardware measurements - Proves genuine Intel SGX/AMD SEV
- Code measurements - Hash of gateway code running in TEE
- Platform certificates - Chain of trust to CPU vendor
- Cryptographic signature - Signed by TEE hardware
Learn About Attestation
Detailed guide on verifying attestation →
Architecture Benefits
1. Privacy by Default
All Providers Protected
All Providers Protected
Whether you use OpenAI, Anthropic, Google, or any of 218+ models, all requests flow through TEE. Privacy is guaranteed regardless of your model choice.
No Provider Access
No Provider Access
Even though we route to providers like OpenAI, your data passes through TEE first. The gateway acts as a privacy shield.
Zero Trust Architecture
Zero Trust Architecture
You don’t need to trust RedPill operators. Trust is rooted in CPU hardware and cryptographic attestation.
2. Compliance & Regulations
TEE protection helps meet regulatory requirements:- HIPAA - Healthcare data protection
- GDPR - European data privacy
- CCPA - California privacy law
- SOC 2 - Security controls
- ISO 27001 - Information security
3. Enterprise Security
Suitable for sensitive workloads:- Legal documents (attorney-client privilege)
- Financial analysis (trading strategies)
- Medical records (patient data)
- Source code analysis (IP protection)
- Personal data processing (PII)
Performance
Near-Native Speed
TEE protection adds minimal overhead:| Operation | Native | With TEE | Overhead |
|---|---|---|---|
| Request latency | 50ms | 51ms | ~2% |
| Throughput | 1000 req/s | 980 req/s | ~2% |
| Token processing | 100 tok/s | 99 tok/s | ~1% |
Scalability
- Horizontal scaling across multiple TEE nodes
- Load balancing within secure enclaves
- No performance degradation at scale
Comparison with Competitors
vs Tinfoil
| Feature | Tinfoil | RedPill |
|---|---|---|
| TEE Models | ✅ Yes (specific models) | ✅ Yes (Phala models) |
| TEE Gateway | ❌ No | ✅ Yes |
| Multi-Provider | ❌ No | ✅ 218+ models |
| All Requests in TEE | ❌ Only specific models | ✅ All requests |
vs OpenRouter
| Feature | OpenRouter | RedPill |
|---|---|---|
| Multi-Provider | ✅ Yes | ✅ Yes (218+ models) |
| TEE Protection | ❌ No | ✅ Full TEE |
| Privacy Guarantees | ❌ None | ✅ Hardware-enforced |
| Attestation | ❌ No | ✅ Cryptographic proof |
vs Direct LLM APIs
| Feature | Direct APIs | RedPill |
|---|---|---|
| Data Privacy | ❌ Provider sees all | ✅ TEE-protected |
| Verifiable | ❌ No | ✅ Cryptographic attestation |
| Multi-Provider | ❌ No | ✅ 218+ models |
| Code Audit | ❌ Closed | ✅ Open source gateway |
Open Source
RedPill’s gateway code is open source, allowing you to:- Audit security implementation
- Verify TEE integration
- Review privacy guarantees
- Build confidence in the system
View on GitHub
Explore the TEE-protected gateway source code →
FAQs
Does TEE slow down my requests?
Does TEE slow down my requests?
No, TEE adds minimal overhead (~1-2%). Most latency comes from the LLM provider itself, not TEE processing.
Can RedPill decrypt my data?
Can RedPill decrypt my data?
No. Data is only decrypted inside the TEE enclave. RedPill operators cannot access the enclave’s memory or decrypt your data.
What if there's a TEE vulnerability?
What if there's a TEE vulnerability?
TEE vulnerabilities are rare and quickly patched by CPU vendors (Intel, AMD). We monitor security advisories and update immediately.
How do I verify TEE execution?
How do I verify TEE execution?
Use the attestation API to get cryptographic proof. See Attestation Guide.
Does this work for all 218 models?
Does this work for all 218 models?
Yes! All requests flow through the TEE-protected gateway, regardless of which model or provider you choose.