vLLM on Inpherio.
Earn from the vLLM you already run. The Inpherio connector finds it automatically and offers the text & chat and embeddings models you choose to developers through one OpenAI-compatible API — at prices you set.
What it is.
vLLM is a high-throughput open-source inference server for large language models, widely used on NVIDIA GPUs, with an OpenAI-compatible API built in.
vLLM loads one model per server and keeps it resident, so it always answers warm. To serve several models, run several vLLM servers on different ports and add the extra ports in your connector settings.
| Serves | Text & chat, Embeddings |
|---|---|
| Detected on | 127.0.0.1, port 8000 or 8001 |
| Text & chat | /v1/chat/completions (also /v1/messages and /api/chat) |
| Embeddings | /v1/embeddings |
| Setup changes needed | None — the connector wraps it |
Set it up in three steps.
Run vLLM
Start vLLM's OpenAI-compatible server with your model, e.g.
vllm serve Qwen/Qwen2.5-7B-Instruct(port 8000 by default).Install the connector
Run the Inpherio install command from your dashboard on the same machine. The connector finds vLLM on port 8000 or 8001 by itself.
Set your prices
Each model is benchmarked on your hardware, then goes live on the marketplace at the platform's default prices. Change any price, or switch a model off, from your dashboard.
The install command is one line on Linux, macOS, Windows, Raspberry Pi or Docker — see the connector guide.
Control it from your dashboard.
What the Inpherio dashboard can do with vLLM on your machine, remotely and securely:
- See live engine detail — loaded models, memory and version
Questions.
Does Inpherio work with vLLM?
Yes. The Inpherio connector detects vLLM running on the same machine automatically and lists its text & chat and embeddings models in your dashboard — the ones you leave switched on are offered on the marketplace, at prices you set. You don't change your vLLM setup.
Do all my vLLM models go on the marketplace?
Only the ones you choose. Every model the connector finds appears in your dashboard, switched on by default; switch any model off (or the whole engine) and it is never routed to, rented or listed — the change applies straight away, no restart. A newly found model also has to pass a benchmark on your machine before anyone can use it.
Which port does Inpherio look for vLLM on?
Port 8000 or 8001 on 127.0.0.1. If yours runs elsewhere, add a port override for vLLM in your connector settings.
Can people reach my vLLM server directly?
No. The connector only talks to your engine on 127.0.0.1 and dials out to Inpherio over an encrypted link — nothing on your machine listens for outside connections.
How are vLLM models priced?
You set the price for each model in pounds — per million tokens for text and embeddings, per hour of audio for speech. Inpherio takes a flat 15% from what you earn, and each payout carries a payout fee of £2 + 1% per payout to a UK bank account, £2 + 2% to any other country. New providers pay no platform fee for their first 12 months or until they have earned £250 fee-free, whichever comes first; the payout fee still applies.