Arcaveli encrypts every AI response with a key only you hold — even a breach of our servers produces zero readable data.
"We use AI every day on client matters. Arcaveli was the first product that let our compliance team actually say yes."
Set up takes under a minute. After that, every response is mathematically inaccessible to us.
An RSA-2048 keypair is generated in your browser. Your private key never leaves your device.
Bring your own Anthropic API key. We never touch your provider billing or rate limits.
Responses are encrypted server-side with your public key. Only your private key can read them.
Consumer AI tools store readable conversation history on shared infrastructure, do not sign Business Associate Agreements, and use prompts to train their models. A single breach exposes every client conversation in plaintext — and there is no remediation that retroactively un-leaks the data.
Arcaveli was built on a single premise: the only secure data is data we cannot read.
{
"user": "Patient presents with chest pain
and shortness of breath. Hx of MI.",
"assistant": "Recommend immediate ECG and
troponin. Consider STEMI protocol..."
}
{
"user": "eyJlbmNyeXB0ZWRLZXkiOiJTM3pE...",
"assistant": "eyJlbmNyeXB0ZWRLZXkiOiJBN3..."
}
Every feature is designed around a single constraint: we never see your data in plaintext.
Responses are encrypted before they touch our database. Plaintext exists only in memory, only for the duration of a request.
Your private key is generated in your browser and never transmitted. We have no recovery mechanism — by design.
Business Associate Agreement available on Starter, included on Business. Audit-ready logging built in.
You retain control of your AI provider relationship. Your keys are encrypted at rest with AES-256-GCM.
Encrypted threads persist across sessions. Resume any past conversation with full context — we still see only ciphertext.
Zero-dependency Node.js SDK. Integrate into existing software in fewer than ten lines of code.
Designed alongside compliance counsel and security engineers from regulated industries.
We sign Business Associate Agreements and store zero readable PHI.
No readable personal data stored. Full data portability and deletion on request.
Audit controls running from day one via Vanta. Report available Q4 2025.
No usage gotchas. Cancel anytime.
No. Every response is encrypted with your RSA public key before it leaves our servers. We never store the matching private key — only you hold it. Even with full database access, our staff sees only ciphertext.
Your existing encrypted conversations become permanently unreadable — that is the cost of true zero-knowledge encryption. We recommend backing up your private key to a password manager and an encrypted offline copy. You can regenerate a new key at any time, but it will not decrypt past conversations.
Yes. BAA is available on all plans and included on Business.
US-based AWS infrastructure (us-east-1) with encryption at rest. EU data residency is available on the Business plan. All inter-service traffic is mutual-TLS encrypted.
All Anthropic Claude models — Opus, Sonnet, and Haiku. OpenAI and Google Gemini support is on our roadmap for Q3.
Anthropic retains conversation data for safety review and may use it to improve their models (depending on your account terms). Arcaveli sits between you and Anthropic: your prompt is forwarded once, the response is immediately encrypted to your key, and nothing is retained in plaintext on our infrastructure.
In transit, yes (TLS). At rest, both your prompt and the assistant response are encrypted with your RSA public key before being written to our database. Plaintext exists only in server memory for the milliseconds required to forward the request to your AI provider.
Install our zero-dependency Node.js SDK (npm install arcaveli-client), generate a keypair, and call client.chat(). Most integrations are under ten lines — see the SDK README for full reference.