OffRouter

active

Private, OpenAI-compatible LLM inference gateway for onchain agents, paid per call in USDC on Arc and Base, USDG on Robinhood Chain. Every model is served in a GPU TEE with a signed receipt. No account, no API key: x402 payment is the auth. Prices per token, unused ceiling refunded.

AIBasex402 v2exactoffrouter.org ↗︎

Settled via Coinbase.

Transactions · 30d
8
Volume · 30d
$0.008
Unique buyers · 30d
8
Uptime · 30d
100.0%
Latency p50
130ms
Reported calls · 30d
11

Endpoints (7 live)

  • POST /v1/chat/completions — Private, OpenAI-compatible chat completions. Every model (Kimi, DeepSeek, Qwen, GLM, gpt-oss and more) runs inside a confidential GPU enclave (Intel TDX + NVIDIA): prompts are encrypted in use, never cached or logged, and each answer carries a signed receipt. Suited to tasks with sensitive data. Priced per call from the input and max_tokens; the unused part is refunded. Model ids: GET /v1/models. (0.001 USDC on Base)
  • POST /v1/rerank — Private rerank (Qwen3 Reranker) inside a confidential GPU enclave: scores documents against a query, best first, for search and retrieval over sensitive documents. The query and documents are never cached or logged. Priced per input token; the unused part is refunded. (0.001 USDC on Base)
  • POST /v1/web/fetch — Private page fetch for agents: reads a web page as clean text from inside a confidential VM (Intel TDX), so the site sees that enclave, never the caller. No cookies or referrer are sent. Requests can be encrypted end to end to the enclave. Flat price per page. (0.002 USDC on Base)
  • POST /v1/web/search — Private web search for agents. The search runs from inside a confidential VM (Intel TDX), so the search engine sees the query and that enclave, never the caller: no IP, account or cookie. Emails, keys and card numbers are taken out of the query first. Requests can be encrypted end to end to the enclave. Each answer carries a signed receipt. Flat price per search. (0.008 USDC on Base)
  • POST /v1/audio/transcriptions — OffRouter private speech to text, pay per call in USDC on Arc and Base, USDG on Robinhood Chain. This amount is one second of audio on voxtral-small-24b; the real amount is quoted per request as a ceiling on the seconds the file can hold, unused part refunded. Send multipart/form-data with model and file; see /openapi.json. (0.001 USDC on Base)
  • POST /v1/embeddings — Private, OpenAI-compatible text embeddings (Qwen3 Embedding, 1024 dimensions) inside a confidential GPU enclave. The text is never cached or logged. Suited to search and retrieval over sensitive documents. Priced per input token; the unused part is refunded. (0.001 USDC on Base)
  • POST /v1/images/generations — Private image generation (FLUX.2 klein) inside a confidential GPU enclave. The prompt is never stored, the image exists only in the response, and each call carries a signed receipt. Priced per image; an image that is not returned is refunded. (0.012 USDC on Base)

First seen · last seen · last active