Host Agreement

In short

• During a rental the whole machine is the renter's. Do not use it, restart it or change it.
• Plan maintenance only for times with no booking.
• A Linux node runs Ubuntu with Secure Boot off, the NVIDIA driver from the Nalvo installer, Docker, the NVIDIA Container Toolkit and the Nalvo egress guard.
• A Windows 11 PC can host too. Rentals run in a WSL2 environment under a separate Windows account that Nalvo's script creates. That account must stay signed in, and the PC is paired from that account only (section 13).
• On a Windows PC, a renter who broke out of a container could reach whatever the rentals account can read. The script keeps that account out of your other drives and folders, but some risk remains (section 13).
• You must never access a renter's container or data.
• Your machine must not be in a country that Mamhol excludes, and you cannot host if the country you declared for your account is excluded. Every node needs a declared country. We check the country you declare for it, your own declared country and, at most steps, a country looked up from your own connection or from the machine's. The check also runs when you list a node again and when a listing re-opens after a pause or a maintenance window. Each time Mamhol switches the check on, or adds a country while it is on, every listed node must confirm its country again through its node agent before it can be booked; the Fleet Manager arranges this automatically (section 2).
• No payouts until a licensed payment provider is live.
• Credits for slow or unavailable GPUs, for your own GPU use on a Windows PC, and host-fault settlements, reduce what you earn.

1. About this agreement

This agreement is between you, as a host, and Mamhol, a company established in Qatar, which runs Nalvo.ai. It adds to the Terms of Service and the Acceptable use policy, which also apply to you.

2. Becoming a host

• Hosting is by invitation. You list a node that is already paired with your Fleet Manager.
• You redeem a one-time enrolment code on the machine itself. A software probe then records evidence about the machine against a single-use challenge from our server.
• An administrator reviews the evidence before the node can be listed. Listings say "Admin-reviewed (software probe)" and "Specs self-reported — not hardware-attested", and nothing stronger.
• You must describe your machine truthfully and must not interfere with the probe.
• Countries. Declare the country where the machine actually is. Every node needs a declared country. The Compute Market is not offered in the countries that Mamhol excludes, and a machine located in one may not be enrolled or listed. Mamhol manages the excluded-country list. You may not host from an excluded country either: if the country you declared for your own account is on the list, then, once the check is switched on, you cannot register a node, create a listing or list a node again, and none of your nodes can be booked, wherever the machines are. When you register a node, we check the node's declared country, your own declared country and a country we look up from the IP address of your own connection; when the node is added from the Fleet Manager, no connection country is looked up and only the declared countries are checked. When the node agent redeems the enrolment code and when it sends its probe evidence, we check the node's declared country and a country we look up from the IP address the agent connects from. When you create a listing, put a node back on the market with "List again", or end a maintenance window early with "End now", we check the node's declared country, your own declared country and your own connection. When a maintenance window ends on time, and when the Fleet Manager resumes a listing it had paused, the node's and your own declared countries are checked again before the listing takes bookings; no connection country is looked up for this. Once the check is switched on, if this does not pass when a window ends, on time or early, the window still ends, but the listing stays closed and the node is out of the market until you use "List again"; if it does not pass when the Fleet Manager resumes a listing, the pause stays. Each booking checks your node's declared country and your own, and whether the node's country is confirmed. Once the check is switched on, a step whose checked country is on the list is refused, and search does not show listings of nodes in excluded countries, of hosts whose declared country is excluded, or of nodes whose country is missing or waiting to be confirmed; before that, in Mamhol's reporting mode, such a step is only recorded. If Mamhol has to roll the Compute Market back to an earlier version, the check may return to reporting mode until Mamhol switches it on again, and some of the checks described here may not run until the current version is back. Once the check is switched on and at least one country is on the list, a node without a declared country cannot be registered, enrolled, listed or booked, and its listings are not shown in search. If a node has no declared country, for example because it was added from the Fleet Manager without one or before the check, you can set it yourself, once, in the Compute Market or through the Fleet Manager. When you do this in the Compute Market, we also check your own connection; through the Fleet Manager, no connection country is looked up. Once the check is switched on and at least one country is on the list, the node then cannot be booked until its node agent has confirmed the country from the machine: the country of the agent's connection is checked then, as at enrolment. The same confirmation is needed for every listed node, and every other node that has already sent its probe evidence, each time Mamhol switches the check on with countries on the list, or adds a country to the list while the check is on. The node agent confirms by sending fresh probe evidence; the Fleet Manager asks it to do this automatically, usually within minutes while the machine is online and paired with it, and you do not need to do anything. A confirmation counts only when a country can be looked up for the machine's connection and neither that country nor the node's or your own declared country is on the list. Until the node is confirmed, it cannot be booked and its listing is not shown in search; a rental already running on it continues unless the law requires an earlier stop. Once a node's country is set, you cannot change it yourself: if the machine moves to another country, tell us at once at hello@nalvo.ai. If your connection or the machine's goes through a VPN, a proxy or a similar service, we see the country of that service instead. How the check works, and what is stored, is described in the Terms of Service, section 3 and the Privacy notice.
• You must have the right to use the machine, the premises, the power and the internet connection for paid hosting.

3. Machine requirements

• Operating system: Ubuntu Server 24.04 LTS, in the release the Nalvo installer uses, or another Ubuntu machine that passes the same checks; or a Windows 11 PC prepared as described in section 13. The GPU, Secure Boot, Docker and system-settings items below are for Ubuntu nodes; section 13 sets out what a Windows PC needs instead.
• GPU and driver: NVIDIA GPUs of the Maxwell generation or newer, not mixing older (Maxwell, Pascal, Volta) and newer generations in one machine, with the NVIDIA driver version the Nalvo installer installs and holds. Do not change the driver while the node is listed, except in a maintenance window.
• Secure Boot off. The Nalvo installer stops if Secure Boot is on, because the NVIDIA driver it installs does not load with Secure Boot on. You accept that turning Secure Boot off removes one protection against low-level malware on your machine.
• Docker: Docker Engine and the NVIDIA Container Toolkit, installed and checked by Nalvo's preparation script, so that a container can see the GPU. A separate Docker storage volume is recommended; the Nalvo installer creates one.
• Egress guard: the Nalvo egress guard, installed by the same preparation script. It puts renters' containers on a separate network that cannot open connections to your local network, your router, the machine itself or other private addresses, or to the public IPv4 address of your internet connection, through which a container could otherwise reach your router and the devices you forward ports to. The node agent keeps that public address up to date for the guard. It is applied again whenever Docker restarts and checked every 5 minutes, and the node agent checks it before every rental start; a rental does not start without it.
• System settings: SSH password login off; automatic upgrades and release upgrades off. Updates are applied deliberately, in a maintenance window.
• Disk: at least 20 GB free on the system disk, plus Docker storage for renters' containers.
• Network: an always-on internet connection. Renters must be able to reach the SSH port of their container at your public address. Rentals use a fixed port range, by default TCP 40000 to 40019; forward that range to the machine on your router. The forwarded ports are then open to the internet. They are for renters' containers only: do not run anything else on them. The node agent itself accepts no incoming connections. A wired connection is recommended.
• Agent: the Nalvo node agent installed, paired and running.

4. A dedicated machine during rentals

From the start hour to the end hour of a rental, including any extension, the whole machine is the renter's. During that time you must not:

• use the GPUs or the machine for anything else, such as games, your own AI work, mining, Nalvo Blocks, or another marketplace such as Salad;
• restart, suspend, switch off, unplug or disconnect the machine;
• stop Docker or the Nalvo node agent;
• change drivers, system packages, firmware or BIOS settings, or network settings;
• run anything that slows the GPUs down.

The Fleet Manager records other programs found using the rented GPUs as contention events, which a renter can use as evidence in a dispute. On a Windows PC the node agent also checks for other GPU use itself during a rental, and that time is credited to the renter automatically (section 13). If you want to use the machine yourself, schedule a maintenance window first (Compute Market, Maintenance section), so that no booking can start.

5. Maintenance windows

• You can schedule a maintenance window only for a time with no booking. The Compute Market refuses a window that overlaps a hold or a booking; schedule it after the booking ends.
• Once a window is scheduled, nobody can book over it. Your listing is paused when the window starts and resumed when it ends, or as soon as you end a window that is under way with "End now", provided the node still passes the country check (section 2). "End now" also checks the country of your own connection. If the check does not pass once it is switched on, the window still ends, but the listing stays closed and the node is out of the Compute Market until you list it again with "List again".
• A window can be scheduled up to 365 days ahead and can last up to 14 days. Choose a reason: power, internet, disk, GPU, software or other.
• You cannot delist a node or put it into maintenance while confirmed bookings remain on it.
• Unplanned downtime during a rental is a host fault (section 7).

6. Renters' containers are off limits

The node agent runs each rental in an isolated container: only the rented GPUs (on a Windows PC, all of its GPUs), limits on memory, processor and processes, all but a few system capabilities removed, no privilege escalation, SSH login by key only, a private workspace volume that is deleted when the rental ends, and a network kept away from your local network by the egress guard.

You must not:

• open a shell in, attach to, inspect, copy, snapshot or change a renter's container, image or volume;
• read or copy renter data from disk, memory, GPU memory or logs;
• capture or inspect a renter's network traffic;
• use a renter's SSH key or connection details for any other purpose;
• change a container's limits, keys or SSH settings;
• tell anyone what you learn about a renter or their work.

The only exception is what the Nalvo node agent does automatically to start, check, extend, stop and delete rentals.

You have physical and administrator access to your machine, so this is a contractual promise rather than a technical barrier. It continues after the rental ends and after you stop hosting. A breach leads to immediate removal from the Compute Market and payouts being held back.

7. Reliability, performance and credits

Heartbeats. During a rental your machine sends a heartbeat about every 60 seconds, with health facts: whether the rented GPUs answer, whether the renter's container runs, whether a GPU reports a slowdown and, on a Windows PC, whether other programs on the PC are using the GPU. Our server's clock decides delivered time. A gap of more than 3 minutes between heartbeats is not delivered time and is not paid.

No-show. If a rental has not started within 10 minutes of the booked start, the renter pays nothing. If your machine reported the failure, your node is also taken out of the market (its listing is paused) and receives a reliability mark.

Lost heartbeat. If our server receives no heartbeat for 3 minutes (or none in the first 2 minutes of a rental), the rental ends as a host fault. The renter pays only for delivered time minus credits. Your node is taken out of the market until you clear the fault and list it again, and it receives a reliability mark.

Stop failure. If the rental container cannot be stopped within 10 minutes of the end, the rental is also a host fault, and the payout for that rental is held until an administrator has reconciled it.

Listing again after a host fault. A host fault (a no-show your machine reported, a lost heartbeat or a stop failure) takes your node out of the Compute Market: its listing stops taking bookings, and it does not come back by itself. Fix the cause first, then list the node again with "List again" next to it under "My nodes" in the Compute Market. "List again" is checked like a new listing, including the country check (section 2). From then on the Fleet Manager's automatic pause applies as usual (see "Offline or not ready while listed" below).

Performance credits. Time when a rented GPU or the renter's container is unavailable is credited to the renter at 100%. Time when a rented GPU reports a hardware slowdown, a thermal slowdown or a power-brake slowdown is credited at 50%. On a Windows PC, time when other programs on the PC use the GPU is also credited at 50% (section 13).

Credits come out of your earnings. Credits reduce the statement total. The platform fee and your share are both calculated from the reduced total. Mamhol does not pay credits separately.

Performance exit. If at least half of the last 30 minutes was degraded or unavailable, and at least 10 minutes were observed, the renter may end the rental early. You are then paid only for delivered time minus credits, and your node receives a reliability mark; its listing stays active.

Reliability on display. Your Fleet Manager shows each node's reliability: the share of the last 7 days covered by real heartbeats. The Compute Market counts reliability marks. We may later require a minimum reliability before a listing can go live, and will announce it in advance.

Offline or not ready while listed. If a listed node with no rental running goes silent for 5 minutes, fails one of its rental readiness checks, or reports that a restart is pending, the Fleet Manager pauses its listing so that no new booking can be made. It resumes the listing when none of these applies any more, unless you paused the listing yourself. Before the listing takes bookings again, the node's declared country and your own are checked (section 2); no connection country is looked up for this. If the check does not pass once it is switched on, the listing stays paused. A running rental is never stopped for this.

8. Prices, fees and statements

• You set the hourly price. Each change creates a new listing version. Bookings keep the version that applied when their hold was placed.
• Renters pay for the whole booked window, except after a host fault or a performance exit, when they pay for delivered time minus credits.
• Mamhol keeps a platform fee from each statement: 10% in the sandbox. When rounding leaves a cent over, it goes to you.
• You are responsible for taxes on your earnings.

9. Payouts

No payouts today. All payments are sandbox payments. Your statements show your share for information only; sandbox amounts are not money owed to you.

When payouts start. Payouts begin only when a licensed payment provider is live, Mamhol's account with it is open, and this agreement has been updated and accepted.

How payouts will work. A rental is paid out after it is settled and reconciled, subject to a holding period so that disputes and chargebacks can arrive first. An open dispute, or an unresolved fault, holds back the payout for the rental concerned.

Refunds after a payout. No payouts are made today, so no refund after a payout can arise.

10. Your other responsibilities

• Keep your machine safe: electricity, cooling, physical security and your own insurance.
• Follow the terms of your internet provider and your premises.
• Keep the node agent's credentials safe, and tell us at once at hello@nalvo.ai if you think your machine has been compromised.
• If you receive an abuse complaint about traffic from your machine, tell us and help us deal with it.
• Do not fake, alter or block heartbeats, health reports or probe evidence.
• Help us resolve disputes about your node.

11. Data about your node

What the node agent and the Compute Market record about your machine is described in the Privacy notice. During a rental, the renter receives your node's public IP address, the container's port and its host-key fingerprint. Listings and renters' rental records show whether your node is a Windows PC or a Linux machine. Renters also see your node's declared country. The country checks on your node, and the flags kept when a check does not pass, are described in the Privacy notice.

12. Pausing, leaving and removal

You can pause or delist a node whenever no confirmed booking remains on it, and uninstall the agent after that. We may pause, delist or revoke a node for a breach of this agreement, repeated faults, or a security risk.

If the country where your node is located is added to the list of excluded countries (section 2), then, once the check is switched on, the node can no longer be enrolled or listed, its listings are not shown in search and new bookings on it are refused. If the country you declared for your own account is added, then, once the check is switched on, you cannot register or list a node or list one again, none of your nodes can be booked, wherever the machines are, and your listings are not shown in search. Whichever country is added while the check is on, every listed node, yours included, must confirm its country again through its node agent before it can be booked again; the Fleet Manager arranges this automatically, usually within minutes (section 2). A rental already running on your node is not cut off in any of these cases: it continues to the end of its booked window unless the law requires an earlier stop. It is flagged for review by an administrator.

13. Windows hosts

A Windows 11 PC can host Compute Market rentals. Renters get the same product as on a Linux node: one isolated container with the GPUs, SSH login by key and a private workspace. The rest of this agreement applies to Windows hosts too; this section adds what is different. Windows hosting is new. The preparation script and the node agent check most requirements on your PC, and a PC that fails a required check cannot be listed. Some things are not yet checked automatically and depend on your setup, in particular whether renters can reach the rental ports from the internet through your router, and whether Windows keeps the PC awake when the agent asks it to.

What your PC needs.

• Windows 11, version 22H2 (build 22621) or later, kept up to date, with virtualization turned on in the BIOS or UEFI settings.
• An NVIDIA GPU that works with CUDA on WSL (GeForce or RTX professional cards of the Pascal generation or newer, in the normal Windows display-driver mode) and NVIDIA's Windows driver version 580 or newer. Do not install a Linux NVIDIA driver in the rental environment.
• All NVIDIA GPUs in the PC are rented together, as one unit. WSL2 cannot give a container only some of them.
• Enough free space on the system drive for the rental environment, which is kept in a virtual disk there with a size cap (see "Preparing the PC").
• A port range for rentals, by default TCP 40000 to 40019, outside the range Windows uses for its own temporary connections.
• One node per PC. A PC can be in the Fleet Manager only once, and a Windows PC that hosts is paired from the dedicated account. If you already paired the PC from your own Windows account, first remove that node in the Fleet Manager, at a time when no confirmed booking remains on it; its agent then uninstalls itself. Then pair the PC from the dedicated account. The first stage of the script warns you when the account you run it from already has the node agent paired.

Preparing the PC. You prepare the PC with Nalvo's script, prepare-rentals.ps1, by following the Windows rentals guide. You can read the script before you run it. It runs in two stages:

• The first stage runs from your own administrator account. It checks the PC, installs or updates WSL (Windows Subsystem for Linux), and creates a separate standard (not administrator) local Windows account for Nalvo, named "nalvo" unless you choose another name, marked as created by the script. If an account with that name already exists and the script did not create it, the script stops, unless you tell it to adopt that account with its -AdoptExistingAccount option. If you do, the script warns you that the files of that account become reachable by a renter who breaks out of a container. Do not adopt your own everyday account or an account that someone else uses. The script adds the dedicated account to the Performance Monitor Users group so that the agent can read GPU usage figures, and keeps it out of your files as described under "Your files" below. It adds a Windows firewall rule for the rental port range (see the next item), and sets sleep and hibernate on mains power to "never", saving your previous settings. The dedicated account has no password unless you give it one (see "Physical access" below), and you can choose to have Windows sign it in automatically when the PC starts.
• The firewall rule lets TCP connections from your network, and through your router from the internet, reach the rental port range in WSL. Windows applies it to WSL as a whole, not only to the dedicated account's WSL: if your own account also runs WSL with mirrored networking and a program there listens on a rental port, that program can be reached in the same way. Do not use the rental ports in your own WSL.
• The second stage runs while you are signed in to the dedicated account, without administrator rights. It sets up that account's WSL with mirrored networking, so that the rental environment uses the PC's own network address, and with memory and processor limits that leave room for you. It installs a Linux environment called NalvoRental from Canonical's Ubuntu 24.04 image, checked against a fixed fingerprint, in a virtual disk on your system drive, and installs Docker, the NVIDIA Container Toolkit and the Nalvo egress guard in it. It then checks that a container can use the GPU, that mirrored networking is active and that the rental ports can be used. When no rental is running, it ends by restarting the dedicated account's WSL (not your own account's), so that the node agent can tell the rental environment apart from other WSL environments on the PC (see "Your own GPU use during a rental").
• The script caps the size of that virtual disk, by default at 256 GB or at the space free on the drive at setup minus 64 GB, whichever is smaller, so that renters cannot fill your drive. Where WSL supports it, the virtual disk also gives space that is no longer used back to Windows; otherwise it keeps the space it has grown to, up to the cap, until you remove the rental environment.
• The second stage's WSL settings belong to the dedicated account. They are designed to leave your own account's WSL, Docker Desktop and other software unchanged. The first stage's changes apply to the whole PC: the WSL update (which briefly stops WSL programs that are running, such as Salad or Docker Desktop), the firewall rule, the sleep settings and the access rules on your drives and folders.
• You set up your router yourself: give the PC a fixed local address and forward the rental port range to it. The script shows the addresses to use. If your internet provider puts your connection behind carrier-grade NAT, renters cannot reach the PC and it cannot host. The forwarded ports are open to the internet; the node agent itself accepts no incoming connections.

Your files. Windows lets every standard account open, and often change, many places outside its own user folder: other drives, folders created directly on the system drive (for example C:\Games), the shared Public folder (C:\Users\Public) and USB drives. A renter who broke out of a container would act with the rights of the dedicated account. So the first stage adds rules that deny the dedicated account access to the top of each of your other drives, to each folder at the top of the system drive other than Windows' own system folders (such as Windows, Program Files, ProgramData and Users), and to the Public folder. It records these rules, and the remove option takes them away again. As the dedicated account, the node agent checks that it cannot open any of these places, nor any other account's user folder, and a PC that fails this check cannot be listed.

• The rules cover the drives and folders that exist when the first stage runs. When you add a drive or create a new folder at the top of the system drive, run the first stage again. Until you do, the check fails and your listing is paused while no rental is running.
• USB and other drives formatted with FAT32 or exFAT have no access rules, so they cannot be protected. Unplug them while your PC hosts. While one is plugged in, the check fails and your listing is paused while no rental is running.
• These rules reduce the risk to your files; they do not remove it. A renter who broke out of a container would still control the rental environment, could open whatever Windows lets every account open (for example the Windows and program folders), and could try to attack Windows itself (see "Risks to your PC"). Keep your personal files in your own user folder or on a drive the rules protect, and keep backups.

Readiness and listing. On a Windows PC the node agent runs under the dedicated account. With its heartbeats it reports yes/no readiness checks: whether it runs under the dedicated account, whether WSL is installed with mirrored networking active, whether Docker and the egress guard work in the rental environment, whether the GPU is visible there, whether the NVIDIA driver is new enough, whether the dedicated account is kept out of your files, whether the agent can read the GPU usage figures, whether Windows has a restart pending, and, for information only, whether the rental ports are the default range. A Windows PC can enrol in the Compute Market only when every one of these checks passes, apart from the pending-restart and port-range facts. While no rental is running, the Fleet Manager pauses a listed PC's listing when it is offline, fails a check or has a restart pending, and resumes it afterwards (section 7).

Keeping the PC available.

• Keep the dedicated account signed in. A locked screen is fine, and so is switching to your own account while the dedicated account stays signed in in the background. Signing the dedicated account out stops rentals. Unless it is signed in again within about 3 minutes, a rental running at the time ends as a host fault (section 7).
• Keep the PC plugged in, and do not put it to sleep, hibernate it, close a laptop lid or shut it down during a rental. The agent asks Windows to stay awake during a rental, but Windows does not let it block a sleep you start yourself. Do not change back the sleep settings the script set.
• Restart for updates only when no rental is running. Set Windows Update's active hours to cover the hours your PC is usually rented: Windows normally restarts for updates outside active hours. While a restart is pending and no rental is running, your listing is paused, so restart soon to have it resumed.
• A restart or sign-out during a rental almost always keeps the PC away for more than 3 minutes. The rental then ends as a host fault: the renter pays only for delivered time minus credits, your node is taken out of the Compute Market with a reliability mark, and once the PC is back and ready you must list it again yourself with "List again" under "My nodes" in the Compute Market (section 7).
• If you chose automatic sign-in, Windows signs the dedicated account in after every restart, so that the agent starts and your PC can take new bookings again without you. It does not save a rental that was running during the restart.
• Microsoft can update WSL itself automatically, and an update can stop the rental environment. If the rental is still running afterwards, the agent starts the renter's container again with its workspace.

Interruptions are credited to the renter. Time lost to a restart, a sign-out, sleep, a WSL update, a power cut or a lost connection is not charged to the renter. While the agent can still report, that time counts as unavailable and is credited at 100%. If your PC stops reporting for 3 minutes, the rental ends as a host fault (section 7). Heartbeats arrive about once a minute. The heartbeat after an interruption, or after a gap the agent did not observe (for example sleep), marks the whole interval since the previous heartbeat as unavailable. A very short interruption that is over before the next heartbeat may not be noticed.

Your own GPU use during a rental. The GPU usually also drives your Windows desktop, but section 4 still applies: during a rental do not use the GPU for anything else. During a rental the agent reads, on your PC, how much programs other than the rental use the NVIDIA GPU, including games, your own AI work, your own WSL environments and other GPU-sharing software such as Salad. When such a program uses more than 15% of a 3D, graphics or compute engine of the GPU for at least half of a minute, that time is counted as degraded and credited to the renter at 50%. Lighter or shorter use, such as ordinary desktop activity, is not detected, and use of the GPU's video encoding and decoding engines (for example to record or stream your screen) is not counted, but section 4 still forbids any other use. If the agent cannot read the GPU usage figures during a rental, that time is also counted as degraded; outside rentals, a PC whose agent cannot read them cannot be listed. To leave out only the rental's own GPU use, the agent must know which of the PC's WSL processes belong to the rental environment. On a PC where another WSL environment runs, for example your own Salad or Docker Desktop, the agent may, once each time it starts, restart the dedicated account's WSL to tell its own environment apart, but only while no rental is running and the preparation script is not running. This does not restart your own account's WSL. While the agent cannot tell them apart during a rental, it counts host GPU use as present: that time is counted as degraded and credited to the renter at 50%, whatever your PC is doing. The credit comes out of your earnings like any other credit, and if at least half of the last 30 minutes was degraded or unavailable the renter may end the rental early. What the agent reads and sends for this check is described in the Privacy notice.

Physical access. Anyone who can sign in to the dedicated account can open the rental environment, including the renter's container and workspace, without any further password. While that account has no password, this is anyone who can use your PC, even when its screen is locked: locking the screen does not stop them. Host only on a PC that only you and people you trust can use. Give the dedicated account a password unless you use the script's automatic sign-in, which works only without a password so that no password is stored on the PC. Renters are told that the host and other people with physical access to the host's PC can reach their rental environment.

The rental environment is off limits. Section 6 applies in full. On a Windows PC this also means you must not open a shell in NalvoRental (for example with wsl.exe), browse or copy its files (for example through \\wsl.localhost), change its settings, the egress guard or the firewall rule, or use the dedicated account for anything other than keeping it signed in and running Nalvo's scripts. Keep the access rules the script set on your drives and folders in place while you host. You are responsible for everyone who can use your PC, including members of your household.

Risks to your PC. Microsoft does not treat WSL as a security boundary, so the renter's container is the main protection between a renter's work and your PC. A renter who broke out of the container would control the rental environment, could reach whatever the dedicated account can open on your PC, could fill the rental environment's virtual disk up to its cap, and could try to attack Windows itself. That is why rentals run under a separate standard account that is kept out of your files as described under "Your files". These measures reduce the risk; they do not remove it.

You stay in control. You can stop hosting whenever no confirmed booking remains (section 12). Running the script with its remove option, in both stages, removes what it set up, such as the rental environment with its virtual disk, the firewall rule and the access rules on your drives and folders, and restores your previous sleep and hibernate settings. Changes you made on your router are yours to undo.