Frequently Asked Questions
Common questions about Harpocrates and confidential AI inference.
How is Harpocrates different from traditional AI APIs?
Traditional AI APIs like OpenAI or Anthropic can see your plaintext prompts and responses. Harpocrates uses Trusted Execution Environments (TEEs) to process your data in hardware-enforced secure enclaves, ensuring that even Harpocrates operators cannot access your sensitive information.
What is a Trusted Execution Environment (TEE)?
A TEE is a secure area within a processor that provides hardware-level isolation for code and data. It ensures that sensitive computations are protected from the host operating system, other applications, and even physical access to the machine. Examples include Intel SGX and AMD SEV-SNP.
Can I use my own models with Harpocrates?
Currently, Harpocrates offers a curated set of models optimized for confidential inference. Enterprise customers can work with us to deploy custom models in secure enclaves. Contact sales for more information.
How much does Harpocrates cost compared to OpenAI?
Pricing is competitive with major AI providers. The privacy guarantees and on-chain transparency justify the premium. Current rates are denominated in ETH on Horizen L3 (for example, 0.0001 ETH per token on testnet); check your dashboard for live pricing and available volume discounts.
What happens if a TEE vulnerability is discovered?
We monitor security advisories from Intel and AMD closely. In the event of a critical vulnerability, we would immediately migrate to patched hardware or alternative TEE technologies. All attestations are versioned so you can verify which enclave version processed your data.
Can Harpocrates see my API keys or passwords in prompts?
No. Your prompts are encrypted client-side before transmission. Only the TEE enclave can decrypt them, and the enclave is designed not to log or export any plaintext data.
How do I verify that my data was processed securely?
Every inference returns a zero-knowledge attestation that you can verify using the SDK's verifyAttestation() method. This cryptographically proves your data was processed in a genuine TEE without tampering.
Why is billing in ETH on Horizen L3?
ZEN is not yet live on the Horizen L3 mainnet, so Harpocrates currently meters and settles usage in ETH on Horizen L3 for transparent, on-chain billing and verifiable receipts.
Is there a free tier or trial?
Yes! New users receive testnet ETH credits to explore the platform. For production use, we offer pay-as-you-go pricing with no minimum commitment.
What's the latency compared to standard AI APIs?
Confidential inference adds ~10-15% latency overhead due to encryption and attestation generation. For most applications, this is acceptable given the privacy benefits. Use llm-secure-fast for lower-latency workloads.
Can I use Harpocrates for real-time applications?
Yes. The llm-secure-fast model is optimized for low latency and suitable for chatbots, autocomplete, and other real-time use cases.
Does Harpocrates support streaming responses?
Streaming is currently in beta. Contact support to enable it for your account. Note that streaming requires additional attestation complexity.
Can I self-host Harpocrates?
Enterprise customers can deploy Harpocrates in their own infrastructure with dedicated TEE hardware. This requires specialized hardware (SGX/SEV-capable servers) and enterprise licensing. Contact sales for details.
What programming languages are supported?
Official SDKs are available for JavaScript/TypeScript and Python. Community SDKs exist for Go, Rust, and Java. The REST API can be used from any language.
How do I report a security vulnerability?
Email security@harpocrates.ai with details. We offer a responsible disclosure program with rewards for valid findings. Please do not disclose vulnerabilities publicly before we've had a chance to address them.