# Aether — inference market for AI agents

> Buy LLM inference from independent GPU sellers through one OpenAI-compatible API.
> Pay only for tokens delivered. Streams survive seller failures (checkpoint + failover).

Base URL: https://appp.tail1cb552.ts.net/v1

## Buy inference (agents)
1. Register: POST https://appp.tail1cb552.ts.net/v1/agents  {"name": "<agent name>", "scopes": ["buy_inference"]}
   -> returns api_key (shown once; store it).
2. Fund your balance with USDC on Ethereum (chain id 1) to the treasury 0x1454Ad4A90ce0c76b70b61004e0a50E6bA33e36A, sent from a wallet you have linked.
   - Link your wallet: GET https://appp.tail1cb552.ts.net/v1/billing/crypto/link-challenge?address=<0x...>
     sign the returned "message" with that wallet (EIP-191 personal_sign), then
     POST https://appp.tail1cb552.ts.net/v1/billing/crypto/wallets {"address": ..., "nonce": ..., "signature": ...}
   - Send the token from that wallet. It is credited after 64 confirmations.
   - Check: GET https://appp.tail1cb552.ts.net/v1/agents/me  (Authorization: Bearer <api_key>)
3. Call any OpenAI-style client with base_url=https://appp.tail1cb552.ts.net/v1 and api_key=<api_key>:
   POST https://appp.tail1cb552.ts.net/v1/chat/completions {"model": "<instrument>", "messages": [...], "max_tokens": 256,
     "max_price_usd_per_mtok": "0.60"}
   Available models and live prices: GET https://appp.tail1cb552.ts.net/v1/models . Price a job first: GET https://appp.tail1cb552.ts.net/v1/quote/<model>?tokens=1000

## Sell inference (GPU owners)
Register with scope "sell_compute", then run app/seller_gateway.py next to your own vLLM/SGLang/TGI server.
It keeps your capacity on the order book and gets paid per delivered token (minus a 5% clearing fee).
Withdraw: POST https://appp.tail1cb552.ts.net/v1/billing/crypto/withdrawals {"amount_usd": "25.00", "to_address": "<your linked wallet>"}

## Reference
OpenAPI: https://appp.tail1cb552.ts.net/openapi.json   Interactive docs: https://appp.tail1cb552.ts.net/docs