Home›Become a Host›FAQ
Thirteen questions, short answers. Some of them answer "no" or "we don't know"; we wrote those down too. Because the product is in Phase 0, many of these items are not working today — we say which ones in every answer.
If your question isn't answered here, write to us from the contact page and we'll add it.
Host section
Frequently asked
No — and that is not a preference, it is the operating condition of the system. The workload runs isolated inside a container, persistent data sits on an encrypted volume, and the customer's secrets (API keys, environment variables) are encrypted at the application layer and never touch the host disk. You supply the machine and the power; you don't look inside. If you could reverse that, the customer would never come to this marketplace — which makes this rule a condition of your income too. Architectural detail: Security.
Whenever you want. The clean way is maintenance mode: the machine stops accepting new jobs, frees up when the pod on it finishes and drops off the list — nothing is written to your scorecard. If you're in a hurry you can also pull the machine out from under a running job; the job moves from its checkpoint to another host and the customer doesn't pay for those seconds. That second path works, but it isn't free: an emergency cut is written to the uptime and job-acceptance metrics. There is no commitment period, exit penalty or lock-in.
There are three options. (1) You define a weekly schedule — for example, on weekdays the machine takes no jobs between 19:00 and 24:00; the scheduler never places a job that would run into that window, and this does not lower your scorecard. (2) You put it into maintenance mode there and then, and it frees up when the current job ends. (3) If you are hosting on Windows through WSL2, using and renting the machine at the same time is possible, but sharing resources lowers benchmark consistency — which is why those machines are placed in the short-job pool. Details: Requirements.
You do. Apart from the commission we cover no item at all: electricity, internet, cooling and hardware depreciation are on the host side. That is why the calculations on the earnings page take the 15% commission off the gross first, then electricity. Without your own kWh price and the average draw of your cards under load, you cannot know your net earnings. Formula and worked tables: Earnings.
The honest answer: we don't know, because nobody can guarantee utilization. Your earnings depend on your price, your card, your region, your scorecard and the demand that month. We have no measured host revenue either — the host network is not live yet. The tables on the earnings page are illustrative calculations for three configurations and three utilization scenarios; run them once against your own electricity price, then decide. If someone promises you a certain number of hours per month, be careful.
We set no limit. Each machine enrolls separately, is priced separately and keeps its own scorecard; they all roll up into a single host account. One machine's low score does not directly penalize the placement of the others, but earnings and payouts merge into one period. If you have a facility with dozens of machines, the Verified tier was written for you.
If you have a metered line, be careful. Model and dataset downloads can run to tens of gigabytes; a busy month means terabytes of traffic. Overage is written to your bill, and if your line is throttled your network component drops — and that is 15% of the scorecard. If you don't have an unlimited line, we suggest keeping your price measured and starting small: measure the first month's traffic, then expand. Target values: download at least 100 Mbit/s, upload at least 20 Mbit/s.
The design goal is to be small: a single static Go binary, a small slice of one core when idle and a few hundred megabytes of memory; on disk, the binary, the logs and the container image cache. Telemetry traffic is continuous but small — the real bandwidth is spent by the workload's own downloads and uploads. These are targets; the way to give measured numbers runs through the agent going live (Phase 1).
You put the machine into maintenance mode and uninstall the agent once it frees up. There is no contract term, notice period or exit fee. What you have accrued is paid in the next payout period — removing the machine does not reduce what you are owed. Your enrollment token and machine registration are deleted from your account; your scorecard history will have reset if you come back, because measurement is kept in a 30-day rolling window.
The weekly period closes every Friday; commission is deducted and the remainder is transferred to your identity-verified account. Hosts in Türkiye are paid by domestic bank transfer, hosts elsewhere through Stripe Connect. You download a line-by-line statement of each period from the panel. On tax we make no promises: your accountant tells you which document you have to issue; our part is to make the receipt/invoice process clear and give you a document for every payout. The payout infrastructure goes live in Phase 1.
You narrow gradually; you don't fall off the list all at once. First you drop out of long and critical jobs, then you fall back in the general pool; below a certain threshold only short, interruptible jobs are routed to you. The panel shows which component fell, which event caused it and the 30-day trend; the warning comes before delisting. Because the score is computed over a rolling window, it starts climbing back the day you fix the cause. Concrete fixes: Scorecard.
Yes, but with limits. The NVIDIA driver stays on Windows and the agent runs in the Linux distribution inside WSL2. Windows updates restart the machine, the WSL layer adds overhead in I/O and pinned memory, and P2P/NVLink should not be expected on a multi-card machine. That is why Windows machines are placed on short, interruptible jobs and rank behind Linux machines in placement. If you want serious earnings we suggest booting Linux from a second disk — same hardware, first-class support. macOS is not supported. With the full reasoning: Setup and OS matrix.
No. The marketplace works by the same rule in both directions: the customer does not pay for seconds that don't work, and the host does not earn from idle seconds. There is no reservation fee, availability premium or floor payment. Two things raise your utilization — your price and your scorecard.
Get started
First see which class your machine's operating system falls into, then clear the hardware thresholds, then work out the earnings with your own electricity price. If all three hold, you'll be ready when host registration opens.
Host registration opens in Phase 1. ; the numbers and examples are illustrative.