← All legal documents

Version 1.0.0 • effective Fri May 15 2026 00:00:00 GMT+0000 (Coordinated Universal Time)

NAL Net Addendum

This addendum applies when you participate in NAL Net (also known as HIVE) — the community-powered GPU marketplace where contributors run inference and generation workers on their own hardware, in exchange for credits paid by job requestors.

NAL Net has two roles:

  • Requestor — you submit jobs (text-to-video, image generation, music) and pay credits to have them processed.
  • Worker (contributor) — you run a NAL Net worker (a wan2gp runtime adapter) on your own GPU and earn credits when your worker processes jobs.

This addendum is part of the Terms of Service and incorporates the AUP, Privacy Policy, and Content License.

Effective date: 2026-05-15


1. Community-Only Network

NAL does not, and will not, operate NAL-hosted workers within NAL Net. Every NAL Net worker is a community contributor running a wan2gp runtime adapter on their own hardware. This is intentional: NAL Net is a marketplace, not a hosted-inference service in disguise.

If you need guaranteed-availability hosted inference, use a hosted endpoint where the model picker offers one (Runware, etc.). NAL Net trades guaranteed availability for community capacity.

2. Requestors — Submitting Jobs

⚠️ Third-party visibility warning. When you submit a job through NAL Net (the default routing for Studio, Music, and selected BRAIN jobs), your prompt and the model's output are visible in plaintext to a third party — a community member running a worker on their own hardware. NAL contractually forbids that worker from retaining, training on, or commercially using your material (Section 3.3.2), and enforces those obligations through monitoring and deactivation — but NAL cannot physically prevent a determined worker from exfiltrating data they receive.

If your prompt or expected output contains personal data of others, trade secrets, confidential business material, or anything you would not show to a stranger, select "hosted endpoint only" in the model picker to exclude community workers. The hosted alternative is more expensive.

2.1. Routing and pricing

2.1.1. When you submit a job that NAL Net can process, the job is dispatched to a worker that has registered to handle the model and parameters you selected.

2.1.2. Pricing is shown before you submit, per the Pricing Schedule. The price reflects what NAL will charge you — the split between worker payout and NAL platform fee is internal.

2.1.3. Failover. If the assigned worker times out, fails, or is unavailable, the job retries on another worker or on a hosted fallback (where the model has one). You are not charged twice. See the audit trail of any single job in your account for the worker chain.

2.2. What the worker sees

2.2.1. The worker that processes your job sees:

  • your prompt (text, image, audio inputs),
  • the model selection and parameters,
  • the output the model produces.

2.2.2. The worker does not see:

  • your NAL account ID or username,
  • your email or other profile data,
  • your wallet balance or payment history,
  • any other job you have submitted before or after this one.

The dispatcher (nal-net) strips identity before forwarding the job.

2.3. Sensitive jobs

2.3.1. Because the worker sees prompts and outputs, do not submit through NAL Net any job whose prompt or output contains:

  • personal data of identifiable individuals,
  • confidential business information,
  • material covered by a non-disclosure agreement,
  • content you would not show to a stranger.

2.3.2. Studio, BRAIN, and nal-music model pickers offer a "hosted endpoint only" option for sensitive jobs. Selecting that option excludes community workers and accepts the higher price.

2.4. Output rights

Output generated on a community worker belongs to you on the same terms as output generated on a hosted endpoint (see Section 5 of the Content License). Community workers do not acquire any rights in the output they produce; their compensation is the credit payout.

3. Workers — Running a NAL Net Worker

3.1. Eligibility

3.1.1. To run a NAL Net worker you must:

  • be at least 18 years old;
  • be the owner of the hardware (or have explicit authority from the owner) running the worker;
  • not be acting on behalf of a cloud-hosting provider sub-renting capacity (NAL Net is for community-owned hardware);
  • comply with all laws of your country of residence regarding running inference workloads for compensation, including tax law.

3.1.2. NAL may decline registration or deactivate a worker at any time, with or without cause, including for performance, security, or abuse reasons.

3.2. Worker software

3.2.1. The NAL Net worker software is the wan2gp runtime with runtime adapters. NAL publishes the worker software with installation instructions at the NAL Net documentation site.

3.2.2. Worker software is provided "as is" with no warranty. You run it at your own risk on your hardware.

3.2.3. You agree to install promptly security-critical updates we publish. Workers running outdated versions may be deactivated.

3.2.4. You may stop running the worker at any time. Stopping does not require notice. Jobs in progress at the time of stop may fail and be retried on another worker; partial work on those jobs is not paid.

3.3. What you process

3.3.1. You do not select which jobs route to your worker. The dispatcher assigns jobs based on the model, parameters, and worker capacity.

3.3.2. You will see prompts and outputs. You agree, with respect to every prompt, reference input, intermediate state, and output that passes through your worker:

  • no commercial use of any kind. You may not sell, license, publish, broadcast, redistribute, or otherwise commercialise the prompts or outputs. You may not derive value from them other than the credit payout NAL pays you for processing the job.
  • no retention. Delete prompts, reference inputs, intermediate state, and outputs from your worker as soon as the job is dispatched back to NAL. The only data you may keep is the job-completion record (job ID, duration, your payout amount) for your own accounting.
  • no copying or redistribution. You may not copy job material to any other system, account, archive, or person. Logging the contents of prompts or outputs to external loggers (cloud logging, monitoring SaaS, your own backups) is forbidden.
  • no training. You may not use prompts or outputs to train, fine- tune, evaluate, distil, benchmark, or otherwise inform any AI model — yours or anyone else's.
  • no analytics or aggregation. You may not analyse prompts or outputs in aggregate to derive trends, patterns, or business intelligence, even in de-identified form.
  • no re-identification. You may not attempt to identify the requestor or any person mentioned in a prompt, including by correlating prompts across jobs, cross-referencing with external data, or social-engineering the platform.
  • no internal use of model outputs. You may not personally use an output that your worker produced for another user. Even though the output passed through your hardware, it belongs to the requestor, not to you.
  • report concerning content via the worker software's report function (see Section 3.3.3) when you encounter prompts or outputs that appear to violate the AUP.

3.3.3. Reporting concerning content. If a job prompt or output appears to violate the AUP (CSAM, terror content, etc.), report it via the worker software's report function. NAL will preserve evidence as required by law and may report to authorities.

3.4. Payouts

3.4.1. Earnings are credited to your NAL wallet per completed job at the worker-payout rate published in your worker dashboard. The rate is a function of the job price (Section 2.1.2) less the NAL platform fee.

3.4.2. Reputation and reliability. Worker payout multipliers are weighted by your worker reputation (heartbeat consistency, completion rate, monotonicity of work). Workers that flap on/off, miss heartbeats, or fail jobs are paid at a reduced rate or deactivated.

3.4.3. Withdrawal. Earned credits are withdrawable subject to Section 4 of the Payments and Refunds Policy, including the 14-day clearing hold and KYC thresholds.

3.4.4. Taxes. Worker earnings are taxable income in most jurisdictions. NAL does not withhold tax. You are responsible for declaring and paying tax on what you earn. Where required by tax law, NAL will issue an annual statement of earnings to workers exceeding the reporting threshold.

3.5. Worker conduct

3.5.1. You agree not to:

  • modify the worker software to bypass auth, monotonicity, or audit controls;
  • run a worker on stolen, unauthorised, or compromised hardware;
  • submit fabricated job-completion claims;
  • run multiple workers on the same physical machine to inflate capacity reporting;
  • use NAL Net to launder funds.

3.5.2. Workers found violating Section 3.5.1 are deactivated and forfeit unpaid earnings; serious violations may lead to account termination and (where applicable) referral to law enforcement.

3.6. Termination of worker participation

3.6.1. You may withdraw as a worker at any time. After withdrawal, any cleared earnings remain in your wallet and remain withdrawable.

3.6.2. NAL may deactivate a worker for the reasons in Section 3.1.2 or 3.5. Deactivation does not forfeit cleared earnings unless Section 3.5 applies.

4. Independence

4.1. Running a NAL Net worker does not make you an employee, agent, partner, or representative of NAL. You are an independent contributor providing computing capacity.

4.2. You are responsible for:

  • your own taxes and social-insurance obligations;
  • the electricity, internet, and hardware costs of running the worker;
  • any regulatory licences required in your jurisdiction to run inference for compensation.

5. Liability

The general limitations of liability in Section 13 of the Terms of Service apply, including the cap on NAL's aggregate liability. Specifically:

  • NAL is not liable for damage to your hardware caused by running the worker, electricity-cost overruns, or business interruption.
  • You are not liable to other users for the output of a model routed through your worker — you process the job, you do not warrant the output.