# OmniProvider (MPP + x402) - https://pay.omnidev.tech

Pay-per-call chat completions and cited briefs, served via an upstream provider. Charged only after the result is produced.
MPP: Tempo charge (pull), tempo-mainnet chain 4217, USDC.e, on /v1/*.

POST https://pay.omnidev.tech/v1/chat/nano  USD 0.001/call  <= 1000 in / 256 out tokens
POST https://pay.omnidev.tech/v1/chat/small  USD 0.002/call  <= 8000 in / 1024 out tokens
POST https://pay.omnidev.tech/v1/brief  USD 0.002/call  cited brief from the caller's quotes: {"query":"...","citations":[{"url":"https://...","quote":"..."}]}
x402 (v2 exact, Base USDC, eip155:8453): the same services and prices on https://pay.omnidev.tech/x402/chat-<tier> and /x402/brief
POST https://pay.omnidev.tech/x402/chat-nano  USD 0.001/call
POST https://pay.omnidev.tech/x402/chat-small  USD 0.002/call
POST https://pay.omnidev.tech/x402/brief  USD 0.002/call
One payment method per URL: /v1/* = MPP only, /x402/* = x402 only (the other method there -> 400, not charged).

Body: {"messages":[{"role":"user","content":"..."}], "max_tokens": 128}
An unpaid request returns 402 with the payment challenge for that URL (WWW-Authenticate on /v1, PAYMENT-REQUIRED on /x402).

Paid requests carry the header X-Omni-Client-Nonce: <22-128 base64url chars> (required on paid requests).
  - The nonce is random, generated by the client before paying, and unique per payment; the client keeps it.
  - A missing or malformed nonce returns 400 client_nonce_required / bad_client_nonce, and the request is not charged.
  - MPP / x402 clients send it as an extra header (examples below); a client that cannot add headers receives 400, not charged.
  - The same credential with the same nonce returns the stored result of that payment, without a second charge.
Nonce example:  N=$(openssl rand -base64 24 | tr '+/' '-_' | tr -d '=')
mppx:       npx mppx --method POST -H 'content-type: application/json' -H "X-Omni-Client-Nonce: $N" -d '{"prompt":"hi"}' https://pay.omnidev.tech/v1/chat/nano
agentcash:  npx agentcash fetch https://pay.omnidev.tech/v1/chat/nano --method POST --header 'content-type: application/json' --header "X-Omni-Client-Nonce: $N" --body '{"prompt":"hi"}' --payment-protocol mpp
x402:       npx agentcash fetch https://pay.omnidev.tech/x402/chat-nano --method POST --header 'content-type: application/json' --header "X-Omni-Client-Nonce: $N" --body '{"prompt":"hi"}' --payment-protocol x402
  (agentcash documents the flag as --header; mppx accepts -H / --header)
Payment is settled only after the completion is produced (settle-after-serve).

Docs: https://pay.omnidev.tech/openapi.json  Terms (draft): https://pay.omnidev.tech/terms  Privacy (draft): https://pay.omnidev.tech/privacy

## p9 catalog (OpenAI-compatible) - p9.12

GET  https://pay.omnidev.tech/v1/models   price per Mtok in/out, context, rails per model
POST https://pay.omnidev.tech/v1/chat/completions   {"model":"<id>","messages":[...],"max_tokens":256}
  MPP tempo charge: 402 WWW-Authenticate: Payment quotes a fixed amount (min $0.0008). /v1 is MPP-only.
GET  https://pay.omnidev.tech/v1/packages   prepaid token packs (OPEN-*) and frontier credit; pay once, then
  Authorization: Bearer sk-omni-...   (or wallet-signed: X-Omni-Wallet / X-Omni-Wallet-Ts / X-Omni-Wallet-Sig)

Client nonce (mandatory on paid requests): X-Omni-Client-Nonce, random base64url, 22-128 chars, new per payment.
  A paid request without it is answered 400 client_nonce_required and is not charged; unpaid price checks omit it.
  The same credential with the same nonce is redeliverable after settlement; a reused nonce weakens that protection.
  NONCE=$(python3 -c 'import secrets; print(secrets.token_urlsafe(32))')
  curl -sS -X POST https://pay.omnidev.tech/v1/chat/completions -H 'Content-Type: application/json' -H "X-Omni-Client-Nonce: $NONCE" -H 'Authorization: Payment <MPP credential>' -d @body.json

Models:
  kimi-k3  in $0.08/Mtok  out $0.24/Mtok  ctx 1048576
  deepseek-v4.1-flash  in $0.03601/Mtok  out $0.036/Mtok  ctx 5000000
  glm-5.3  in $0.4/Mtok  out $0.2/Mtok  ctx 5000000
  llama-3.1-8b  in $0.06/Mtok  out $0.12/Mtok  ctx 16384
  gpt-oss-120b  in $0.12/Mtok  out $0.5/Mtok  ctx 122880
  gpt-6-luna  in $0.15/Mtok  out $0.75/Mtok  ctx 1000000
  claude-haiku-5.5  in $0.15/Mtok  out $0.75/Mtok  ctx 100000
  gpt-6-sol  in $2.4/Mtok  out $12.0/Mtok  ctx 1000000
  claude-sonnet-5.5  in $2.4/Mtok  out $12.0/Mtok  ctx 1000000
  gemini-3.1-pro  in $2.4/Mtok  out $14.4/Mtok  ctx 200000
  grok-4.7  in $2.4/Mtok  out $7.2/Mtok  ctx 500000
  gpt-6-astra  in $12.0/Mtok  out $60.0/Mtok  ctx 1000000
No streaming, no tools. The charge occurs only after the completion is produced. Served via an upstream provider; while the upstream is down every paid route answers 503 upstream_unavailable (charged:false) before quoting.