FAQ

B3IQ is a neocloud: dedicated GPU machines, bought through B3IQ or enrolled from your own hardware, that serve your inference free and unmetered through an OpenAI-compatible gateway. The host agent gives a machine a node identity, local OpenAI-compatible endpoints, runtime and benchmark probing, signed heartbeats, and receipts, and the control plane binds it to a registry so your own API keys and CLI can reach it. Earning is optional, and it is a property of the machine's hosting arrangement rather than of the requests it serves: B3IQ sells the machine's capacity by the hour, and the machine's share of that accrues to its deal ledger.

The local API runs on your own machine. The host agent serves an OpenAI-compatible endpoint at http://127.0.0.1:8831, backed by whatever runtime and model you have installed. No routing, no billing. Your prompts and outputs stay local unless you send them to a routed workflow.

The hosted gateway at https://b3iq.org/v1/api is a network endpoint. It routes your requests across the machines on your own account by health and readiness, enforces customer policy, and meters usage. Inference is free. You call it like any OpenAI-compatible API with a gateway key.

There's no approval step. Joining the fleet is invite/manual onboarding: a new machine enrolls with an onboarding code the B3IQ team issues, and a machine that was pre-onboarded on your behalf is linked to your account by redeeming a claim code in the dashboard. Once a machine is onboarded and claimed, it serves your own inference immediately. There's no admin sign-off in between, and no separate review gates earning either.

B3IQ installs and supervises Ollama, llama.cpp, LocalAI, and Xinference, and it can also point at a runtime you provisioned yourself as long as it speaks an OpenAI-compatible API (vLLM, SGLang, TensorRT-LLM, and others). Machines run Linux. The setup wizard probes the machine and recommends a runtime; the host agent verifies it is reachable before a model profile is marked route-ready.

Yes. The b3iq CLI signs in with b3iq login (a browser device flow) or a personal access token from Settings → CLI tokens, then lets you list machines, install or remove models, restart runtimes, tail logs, and open an SSH session, all without touching a ProxyCommand. Each machine also has a native SSH lane you can enable from its dashboard page, and organizations can set fleet-wide SSH keys that reconcile onto every machine they own.

Transport is encrypted from public surfaces and between nodes and the control plane, and the control plane does not retain the prompts or outputs of routed gateway requests. But be clear-eyed about the trust boundary: when work runs on a node, the node operator can see the plaintext prompt and output for that request, because inference happens in process on their machine.

Verified-confidential / TEE execution (where the operator could not read request content) is future work, not something B3IQ provides today. Do not treat routed inference as confidential against the node operator.

Earning is a property of the machine's hosting arrangement, not of an individual request. An earning machine takes a revenue share on what it makes under B3IQ's offtake arrangements; the share accrues to that machine's deal ledger and is paid in USD. A served request is metered and receipted, but it does not itself pay you.

To be available for that work, the machine has to be claimed to your account and route-ready: online, in its default bare-metal operating mode rather than Earn mode, with the model installed, benchmark-observed, and queue-healthy. There's no separate approval step, and an installed model alone is not enough.

Inference is free. Buy (or bring) a machine, claim it to your account, and your API keys route to it at no per-request charge: there are no credits, no prepaid balance, and no top-ups. Usage is metered per key and per account so you can see what ran, and accounts carry an optional daily token quota that defaults to uncapped. Serving your own requests is not what pays you — a request is metered and receipted, but the machine earns on the capacity B3IQ sells by the hour under your hosting arrangement. See the earnings page.

Open a ticket from the dashboard's Support tab. Replies from the B3IQ team land on the same thread. B3IQ also emails a machine's owner automatically if it goes offline, its GPU runs hot, or its disk fills up, so you don't have to watch the dashboard for early warning signs.

No. B3IQ is in pilot, and we report status honestly. The hosted gateway, receipts, and payout settlement are live for pilot use, but the network is still maturing. Check the current-status page for live evidence and launch gates before you build on it.