Inference that attests itself.
Attesta runs models inside confidential containers — hardware-encrypted enclaves where inference is sealed from the host. Every response returns a signed attestation you can verify yourself.
Built on hardware roots of trust
From request to proof in three guarantees.
Every inference call passes through a chain of hardware-enforced controls. Nothing runs until the environment proves what it is.
Confidential containers
Your model runs inside a hardware-isolated TEE. RAM is encrypted with a key held only by the CPU — the host OS, hypervisor, and our own operators can never read it.
Remote attestation
Before a single byte flows, the enclave produces a signed quote — proving its identity and the exact code measurement, attested all the way down to the silicon vendor's root key.
Verifiable inference
Every response ships with an attestation token bound to that exact request. Verify the environment cryptographically — in your own client, on every call, with no trust in us required.
Drop-in private inference, without the trade-offs.
The performance and ergonomics of a hosted API — with guarantees you usually only get on your own hardware.
OpenAI-compatible
Point your existing SDK at our endpoint and ship. Same request shape, same streaming — plus an attestation header on every response.
Zero data retention
Prompts, completions, and embeddings live only in encrypted memory for the life of the request. Nothing is logged, cached, or used for training. Ever.
Open & frontier models
Llama, Mistral, Qwen, DeepSeek and more — or bring your own weights into a sealed container we never inspect.
Hardware root of trust
Trust chains terminate at Intel and AMD signing keys — not at a TLS cert or a promise. Verification is rooted in silicon.
Zero chat retention
Conversations are encrypted in your browser with a device-bound key and never stored on our servers. Prompts are processed in the enclave, not warehoused.
Audit-ready logs
Export tamper-evident attestation records for every call — the compliance evidence your security team actually asks for.
Change the base URL.
Keep your stack.
If you can call OpenAI, you can call Attesta. The difference arrives in the response — a verifiable attestation, attached automatically.
One-line migration
Swap the base URL and key. No SDK changes, no new client.
Verify in-band
Pull response.attestation and check it against the silicon root — in three lines.
from openai import OpenAIfrom attesta import verify client = OpenAI( base_url="https://api.attesta.sh/v1", api_key="sk-att-…",) resp = client.chat.completions.create( model="llama-3.3-70b", messages=[{"role": "user", "content": "Summarize…"}],) # prove the environment that produced itverify(resp.attestation) # → AttestationValid ✓Don't trust the platform.
Verify the silicon.
Most “private” APIs ask for your trust. Attesta hands you a cryptographic proof instead — one your own code can check, independent of us.
- Measured environment. The exact container image and runtime are hashed and signed at boot.
- Rooted in hardware. Signatures chain to Intel/AMD vendor keys — not a certificate we control.
- Bound per response. Each token ties the proof to one specific request, so it can't be replayed.
Built for teams that can't afford to trust.
“Finally, AI inference I can show to legal.”
Head of Engineering
FinTech startup
“We passed our SOC 2 audit specifically because of Attesta’s attestation receipts.”
CISO
Healthcare SaaS
“Dropped in as an OpenAI replacement in 2 hours. The attestation just works.”
Staff Engineer
AI Platform team
A threat model that assumes the worst.
Attesta is designed so that even a fully compromised host — or a malicious insider at Attesta — still cannot read your data.
SOC 2 Type II
Independently audited controls across security, availability, and confidentiality.
HIPAA-ready
Process PHI with BAAs in place and no plaintext ever leaving the enclave.
GDPR & data residency
Pin inference to EU regions; no cross-border movement of prompts or outputs.
Open verification
Published build measurements and an open-source verifier — audit us yourself.
Private inference, predictably priced.
Every plan runs in confidential containers with full attestation. No data retention on any tier.
- 100K tokens / month
- 10 requests / minute
- Per-response attestation receipts
- Community support
- Unlimited attested requests
- Frontier + open models
- Bring-your-own-weights into sealed containers
- GPU-accelerated inference (NVIDIA A10)
- Audit-ready attestation logs
- Priority support
- Dedicated single-tenant enclaves
- EU / US data residency pinning
- SOC 2, HIPAA BAAs, custom DPAs
- SSO, SLAs & dedicated support
Start with 100K free tokens. No credit card.
Every request is attested. Every response is provable. Drop in and see.
Free forever · Attestation on every request · OpenAI-compatible