AI inference
An OpenAI-compatible API where every request is paid work the chain mines: the weight matrix products of your tokens become proof-of-work jobs, and each call returns a receipt you can follow to the blocks it secured.
Playground
Ask anythingDecoding is greedy, so the same prompt gives the same answer. Each reply comes with its receipt.
The gateway is offline on this network, so replies will fail until it is back.
Prepaid balance
Not connectedTop up once with SOL, then send requests with no transaction per call.
Paid in SOL over x402 through the storefront. Unused rows are refunded on the chain when the balance closes.
API key
A key bound to your wallet lets any OpenAI client draw on your prepaid balance. Creating one is a wallet signature, not a transaction. The key is shown once.
Models and prices
priced for 64 prompt + 64 output tokens| Model | Price | Served |
|---|
Your receipts
Not connectedConnect a wallet to see a receipt for every call: what you paid, the ticket, the blocks mined on it and its proof.
For agents
from openai import OpenAI
client = OpenAI(base_url="https://api.64-95-12-116.sslip.io/v1/openai", api_key="cadre_sk_...")
r = client.chat.completions.create(
model="MODEL_NAME",
messages=[{"role": "user", "content": "Classify: 'refund my order'"}],
max_tokens=64,
)
print(r.choices[0].message.content)
print(r.cadre) # receipt: request_id, ticket_id, charged_lamports, explorer_urlBase URL
https://api.64-95-12-116.sslip.io/v1/openai
Auth
Bearer key, wallet signature, or x402 per request
Streaming
SSE, usage and receipt in the last chunk
Decoding
greedy (temperature and top_p accepted, documented)
Receipt headers
X-Cadre-Request, X-Cadre-Ticket, X-Cadre-Charged-Lamports
Short-context and batch work fits best: classification, extraction, judging, tagging. A request draws rows from your balance with no transaction on the request path; the gateway settles rounds on the chain.