Frequently asked questions.

Straight answers, including the ones where the honest answer is “we don’t know yet”.

Is Nodefield live yet?

No. The network is pre-launch. There are 0 live workers, no models serving live traffic, and the API is in demo mode on localhost. The waitlist, the demo API and this site are what exist today.

How much can I earn from my GPU?

We do not know yet, and we will not guess. Earnings depend on real developer demand, your hardware and uptime, and prices that are still being validated against measured costs. The proposed split is 80% provider / 20% platform, pending validation. A machine that is merely connected earns nothing.

Will the agent run arbitrary code on my computer?

No. The agent runs sandboxed inference workloads only — a model and a prompt under strict resource limits. It never executes arbitrary customer code, never accesses personal files, and never hides what it is doing. You can pause or uninstall it at any time.

Which GPUs work?

Consumer GPUs with enough VRAM for quantised models: broadly, NVIDIA cards with 6 GB+ and AMD cards with 8 GB+ where the runtime supports them. Under 4 GB VRAM is unsupported. See the compatibility table on the providers page — the agent verifies your actual hardware.

Which models will be available?

Launch candidates are Qwen3-4B, Qwen3-8B, Qwen3-14B, Llama-3.1-8B and Mistral-7B, all quantised. All are currently planned, not serving live traffic. Models are added only after they run and are benchmarked on representative hardware.

Is the API OpenAI-compatible?

Yes, by design. POST /v1/chat/completions takes the same shape as OpenAI’s chat completions, so you should be able to swap the base URL to http://localhost:8787 (demo) once the network is live. See the docs page for examples.

How much does it cost?

Nothing today — no billing is active. Proposed launch pricing is Economy $0.15/$0.60, Standard $0.30/$1.20 and Priority $0.60/$2.40 per million input/output tokens, with Batch at 50% off Standard. These are targets pending cost validation, and beta testing is free within resource limits.

Do you split one model across many GPUs?

No. Each request is assigned whole to a single suitable machine. Splitting one model across geographically distributed consumer GPUs is not part of the design.

What data do you collect?

Only what the waitlist asks for: name/username, email, country, hardware details for providers, and intended use for developers. We do not sell personal data. See the privacy policy.

Is this a crypto or blockchain project?

No. Nodefield is infrastructure, not a token or a chain. There is no coin, no mining, and no decentralisation claim beyond the plain fact that compute comes from many independent machines.

When is the beta?

After Phase 2 (first model, benchmarked) and Phase 3 (GPU agent) are done and tested. The private-beta targets are 20 functioning GPU contributors, 10 developer testers, 3 supported models, a working API and public benchmark results. Join the waitlist to be contacted when it opens.

How do I report a bug or suggest a feature?

Join the waitlist and use the community channels once they open, or raise it in the project repositories once they are public. Include steps to reproduce and what you expected to happen.

Still stuck? Join the waitlist and tell us what you need — early questions shape the product.