a two-way radio for GPUs

Tune in to the world's
home GPUs.

Local-AI enthusiasts go on air with their self-hosted models — like ham radio, for LLMs. You tune in and pay per token. One CLI.

macOS · Linux · Windows — no account needed to browse or download a binary →
The whole thing

Run rogerai. Tune in. Get an endpoint.

Browse stations by signal, tune in, and RogerAI hands you a local OpenAI-compatible endpoint + key on 127.0.0.1 — a drop-in for Hermes, your bots, or any tool that speaks OpenAI.

Live from the band

The signal tower.

Every model is a channel. More online providers = stronger signal, faster tokens, lower price. As a channel fills up, signal and quality dip and the price drifts up — supply and demand, in real time.

tuning in to the broker…

live from broker.rogerai.fyi/discover · prices in $ / 1M tokens

a million tokens, typically $0.18 – $0.55 in & out · set by each station · ~150–400 chat replies
your wallet $42.17 prepaid credit · top up with Stripe · debited per token
a typical reply costs $0.0001 24 tokens out @ 0.22 $/M · you watch it tick, live
The idea

Hobbyist GPUs, on the air.

There's a whole community running powerful open models on hardware at home. RogerAI turns that idle capacity into a shared band — discoverable, fair, and dead-simple to join from either side.

go on air

Share a model

Point RogerAI at your local backend (vLLM, llama.cpp, Ollama). Set a call-sign, a price, and the hours you're around. Flip the switch — you're on air.

rogerai share waterfall --rate 0.30
tune in

Use a model

Browse stations by signal, price, and distance. Tune in and RogerAI hands you a local OpenAI endpoint + key — a drop-in for anything that speaks OpenAI.

rogerai connect @nightowl
How it works

Flexible by design. Easy by default.

On air / tune in

Two verbs. Owners broadcast a model; users tune to a channel. Signal bars are tokens per second, so good and bad are obvious at a glance.

Lineage receipts

A gold call-sign ◆ means every token is signed with its lineage — base weights, fine-tunes, and the exact station that aired it.

Your price, your terms

Set per-token rates per model — or share for free. Open only during your hours, cap usage, throttle by time of day. The market clears the rest.

Only the model. Nothing else.

RogerAI bridges the LLM endpoint and only that. No shell, no files, no context, no personal data — ever reachable. The rest of your machine stays yours.

Monetize idle silicon

Your GPU pays rent while you sleep.

Set a fair rate and keep the lion's share of every token you serve. Free is welcome too — plenty of stations run open for the love of the band.

  • ✓ Per-input and per-output pricing, per model
  • ✓ Time-of-use windows & usage caps you control
  • ✓ Provably fair metering — no under-reporting, no over-charging
  • ✓ Revenue share that favors the operator

Pick up the mic.

Install the CLI and you're one command from on air — or tuned in.