Operational trust for AI
Every AI call
signed.
Verifiable.
One OpenAI-compatible endpoint in front of every frontier and open-weight model. Attestic policy-routes, meters real tokens, and writes a hash-chained, Ed25519-signed receipt for every call — provable offline, anytime.
POST /v1/chat/completions
{"model":"gpt-5.5","stream":true}provider → openai model → gpt-5.5 policy → typed-config
…Hello, and welcome to…
ed25519 → a3f7c2… entry → sha256 chain → +1 record
x-attestic-receipt-id: b018e5… x-attestic-entry-hash: 7795fe…
What Attestic actually does
Operational trust — provable, not claimed.
Policy routing
Send any model id, Attestic routes it to the right provider across OpenAI, Anthropic, Google, and 118+ NVIDIA open-weight models. Per-tenant overrides, rate limits, and monthly budgets enforced at the gateway.
Ed25519 receipts
Every call is appended to a hash chain and signed with a persistent Ed25519 key. The receipt, request id, entry hash, signature, Merkle inclusion proof, is verifiable offline against the public key. No access to Attestic required.
Real metering
Token usage comes straight from each provider's stream, never estimated. Cost is computed from a web-verified price table. Unknown models meter as priced:false rather than guess. Usage, cost, and budget status available via API.
One base URL swap
If your app already uses the OpenAI SDK, you're 3 lines away.
from openai import OpenAI
client = OpenAI(
base_url="https://atherion-gateway-924270273440.us-central1.run.app/v1",
api_key="atk_your_key_here",
)
response = client.chat.completions.create(
model="gpt-5.5", # or claude-opus-4-8, gemini-3.1-pro-preview,
messages=[{"role": "user", # meta/llama-3.3-70b-instruct, …
"content": "Explain provenance in one sentence."}]
)
# Every call → signed receipt in x-attestic-receipt-id headerProve what your AI did.
Every call audited, metered, and signed from the first request.