Terms of Service

1. Acceptance of Terms

By accessing or using LLM Prover (“the Service”), you agree to be bound by these Terms of Service.

2. Description of Service

The Service provides LLM benchmarking, evaluation, and compliance tooling as described at llmprover.pysolvr.com. The Service is operated by pysolvr LLC, a Wyoming limited liability company.

3. Access and Credentials

• You are responsible for maintaining the confidentiality of your access credentials.
• You may not share, sell, or transfer your access to third parties.
• Usage is subject to the rate limits and monthly usage limits of your subscription tier.
• We reserve the right to suspend or terminate access immediately, without notice, for abuse or violation of these terms. “Abuse” includes, without limitation, using the Service as a conduit or vector to attack, probe, overload, extract data from, or circumvent the safety controls of any third-party system, model, or provider; and any use prohibited under Section 3a (Acceptable Use).

3a. Acceptable Use

The Service exists to test and measure the quality, behaviour, and compliance of AI models and your own systems. You must not use it as a tool, proxy, or attack vector against anyone else. In particular, you must not use the Service, or your credentials, to:

• attack, probe, penetration-test, overload, degrade, or attempt to extract data or system prompts from any model, endpoint, system, or provider you do not own or are not explicitly authorised to test;
• develop, refine, stage, or launder prompts or payloads intended to cause harm, circumvent a provider’s safety or usage controls for a malicious or unauthorised purpose, or facilitate unlawful activity;
• generate content that is unlawful, or that infringes, defames, or violates the rights of others;
• evade rate limits or usage limits, or route another party’s traffic or workload through your account; or
• otherwise use the Service in a way that breaches an applicable LLM provider’s terms (Section 10), the terms of any endpoint you register (Section 11), or any applicable law.

Security and adversarial testing (red-teaming, jailbreak, prompt-injection, and similar) is permitted only against models, endpoints, or systems you own or are explicitly authorised to test. You are responsible for holding that authorisation.

Where prohibited, abusive, or unauthorised use is identified or reported (including by an affected provider), we will suspend or terminate your access on a provider’s reasonable request, and we will cooperate in good faith with providers in investigating and remediating the issue, including by providing relevant logs and records. We may act immediately and without notice, reserve the right to pursue legal action and recover resulting damages, costs, and expenses, and may cooperate with law enforcement or other authorities.

4. Subscription and Billing

• Paid plans are billed monthly unless otherwise stated.
• You may cancel your subscription at any time. Access continues until the end of your current billing period.
• Subscription fees are billed monthly in advance and are non-refundable. Downgrading or cancelling mid-period does not entitle you to a partial refund. See our Refund Policy for details.
• We reserve the right to change pricing. We will endeavour to provide reasonable notice of material changes.

5. Intellectual Property

• You retain ownership of all data you submit to the Service.
• Output generated by the Service is yours to use without restriction.
• The Service itself, including its code, design, and documentation, remains our property.

6. Limitation of Liability

The Service is provided “as is” without warranty of any kind. We are not liable for any indirect, incidental, or consequential damages arising from your use of the Service.

7. Termination

We may terminate or suspend your access at any time for violation of these terms. You may terminate your account at any time by cancelling your subscription and ceasing use.

8. Changes to Terms

We may update these terms from time to time. Continued use of the Service after changes constitutes acceptance.

9. Contact

For questions about these terms, visit our contact page.

10. Third-Party LLM Providers

When you use the Service, your prompts and context (including fragments of any RAG content you upload) are sent to third-party LLM providers to generate responses. This is the core function of the product.

We are not responsible for how LLM providers handle inference data.

Providers available through the Service are listed on our Models page. You should review each provider’s privacy policy before submitting data. If your data is subject to regulatory requirements (HIPAA, PCI-DSS, GDPR special categories, export controls), you are responsible for ensuring the relevant providers meet those requirements before use.

Our liability is limited strictly to data we store and control. We make no representations about provider data handling and accept no liability for provider data practices.

Provider acceptable-use policies apply to you. Each LLM provider the Service routes to imposes its own acceptable-use policy, usage policy, and terms of service. Those policies are incorporated into this Agreement by reference as they apply to your prompts and usage, and you agree to comply with the acceptable-use policy of every provider your usage reaches, including, as applicable, the policies of OpenAI, Anthropic, xAI, Google, Groq, Together, and any other provider listed on our Models page. Where a provider’s policy is more restrictive than these Terms, the more restrictive rule governs your use of that provider. You are responsible for reviewing those policies; they are available from each provider. Because we access these providers on your behalf, your breach of a provider’s policy is a breach of these Terms, and a provider may throttle, block, suspend, or terminate access, or require us to suspend your access, in response, which we may do without notice.

11. Bring Your Own Endpoint (BYOE)

Pro and above subscribers may register external LLM endpoints (“BYOE endpoints”) for use in benchmarks and evaluations. By registering a BYOE endpoint you agree to the following:

Credential storage. Your endpoint URL and authentication credential are stored encrypted at rest using AWS-managed encryption. Credentials are never logged, never returned in API responses after initial registration, and are only decrypted at the point of making a benchmark call.

Usage scope. We will only call your endpoint to execute benchmark runs you have configured. We will not use your endpoint for any other purpose or share your endpoint details with any third party.

Cost responsibility. Benchmark runs against your BYOE endpoint generate real API calls to your provider. You are solely responsible for all costs incurred on your provider account as a result of benchmark runs. This includes scheduled runs that fire automatically. We have no visibility into your provider’s pricing and accept no liability for charges incurred. You should review your provider’s pricing and set appropriate usage limits on your provider account before registering an endpoint.

Security recommendation. We recommend using a scoped API key for your BYOE endpoint rather than a root or admin credential. Limit the key’s permissions to inference only.

Data flow. Benchmark prompts from your configured benchmarks will be sent to your endpoint. Responses from your endpoint will be stored in our systems and scored. The same data handling terms that apply to managed model responses (Section 10 and our Privacy Policy) apply to BYOE responses.

Removal. You may delete a BYOE endpoint at any time via the dashboard. Deletion removes the stored credential. No further calls will be made to your endpoint after deletion. Existing run records that reference the endpoint are retained per our standard data retention policy.

12. Compliance Evaluation

The Service provides tooling to evaluate LLM outputs against regulatory and compliance criteria (“compliance evaluation”). By using compliance evaluation features you acknowledge the following:

Not legal advice. Compliance evaluation results are automated assessments produced by an AI judge. They are not legal advice, legal opinions, or formal compliance certifications. Results should not be relied upon as a substitute for qualified legal or regulatory counsel.

No guarantee of accuracy. The AI judge evaluates responses against the criteria and regulation text you provide. The quality of results depends entirely on the quality of your compliance store, your rubric criteria, and the regulation text you upload. We make no warranty that results are accurate, complete, or current.

Your responsibility. You are responsible for: (a) ensuring the regulation text in your compliance store is accurate and up to date; (b) interpreting results in the context of your specific regulatory obligations; (c) obtaining qualified legal advice before making compliance decisions based on Service output.

No certification. A passing compliance evaluation result from the Service does not constitute regulatory certification, legal clearance, or any form of official compliance status. Do not represent Service output as a compliance certification to regulators, auditors, or customers.

Data you upload. Regulation documents and compliance criteria you upload are stored and used solely to power your evaluations. They are not shared with other users. Standard data handling terms apply (see Privacy Policy).

13. Acceptable Data

The Service is designed to test prompt quality and model behaviour. It is not designed to process sensitive personal or regulated data.

Accepted: Prompts and system instructions, synthetic or anonymised test cases, internal documents with no PII or regulated data, ground truth answers derived from internal knowledge.

Not accepted: Personal data (names, emails, addresses, IDs), financial data (card numbers, account numbers), health data (medical records, diagnoses), data subject to export controls, customer data of any kind, or any data regulated under HIPAA, PCI-DSS, or GDPR special categories.

You are responsible for ensuring data you submit complies with these restrictions. We reserve the right to suspend accounts where prohibited data is identified or reported.

14. Illustrative Data

Charts, graphs, and data visualisations displayed on the Site or in marketing materials may use illustrative figures. Illustrative figures are representative examples intended to demonstrate product functionality and are not derived from actual benchmark runs. They do not represent the performance, accuracy, cost, or behaviour of any specific AI model or provider. pysolvr LLC makes no warranty that any model will perform as depicted in illustrative examples. Users should run their own benchmarks on their own data to obtain actual performance figures.

15. Programmatic and Agentic Use (API and MCP)

The Service can be driven programmatically through the API and through the Model Context Protocol (MCP) server, including by autonomous AI agents. These interfaces let software, not just a person clicking in the dashboard, initiate cost-generating operations such as comparisons, evaluations, benchmark runs, and scheduled jobs. By using the API or MCP interfaces you acknowledge and accept the following.

Agentic behaviour is non-deterministic and outside our control. When you connect an AI agent (via MCP or your own code) to the Service, that agent is yours. It may call Service tools in ways you did not explicitly anticipate, repeat calls, loop, or act on its own reasoning. AI agents are probabilistic and non-deterministic by nature. We do not control, supervise, or predict your agent’s behaviour, and the recipes and example choreographies we publish are guidance, not guarantees of how any given agent or client will actually behave.

Cost-generating calls are your responsibility. API and MCP calls that run comparisons, evaluations, or benchmarks consume your plan’s usage and may incur charges, including charges on your own third-party or BYOE provider accounts (see Sections 10 and 11). An agent or script acting on your credentials can generate such calls, including repeated or looping calls, without a human approving each one. You are solely responsible for all usage and all costs generated through your credentials, whether initiated by you, your software, or an AI agent acting on your behalf.

You must monitor your own automation. It is your responsibility to supervise, bound, and monitor any agent or program you connect to the Service: set budgets and usage limits on your account and on your provider accounts, use the stop conditions and cost ceilings the recipes provide, and watch for runaway or unintended activity. The guard-rails we build into recipes (iteration caps, cost ceilings, readiness checks, confirmation prompts) are provided to help, but they are implemented by the agent following guidance and are not a substitute for your own monitoring and limits.

Credential control. You are responsible for safeguarding API keys and MCP credentials. Any activity performed with your credentials is treated as authorised by you. Revoke or rotate a credential immediately if you suspect it is misused or if an agent is behaving unexpectedly.

Limitation of liability. To the maximum extent permitted by law, pysolvr LLC is not liable for any usage, charges, data exposure, or damages (direct, indirect, incidental, or consequential) arising from programmatic or agentic use of the Service, including costs generated by autonomous agents, scripts, scheduled jobs, or non-deterministic agent behaviour. This is in addition to, and does not limit, the general limitation of liability in Section 6.

16. Cost Estimates

Cost figures shown by the Service (per-model costs in comparisons, evaluations, and benchmark runs, cost estimators, and value/quality-per-dollar metrics) are estimates calculated from standard published provider pricing. They are a reference point, not a bill.

Your actual costs may differ. Published rates are the baseline list price. Your real cost with a provider can be lower or higher depending on your own pricing tier, committed-use or volume discounts, negotiated rates, promotional credits, region, batching, caching, or changes a provider makes to its pricing. For Bring Your Own Endpoint (BYOE) models, costs are calculated from the rates you supply (or, if you supply none, are shown as unavailable), and the real charges land on your own provider account (see Section 11).

Use estimates for comparison, not accounting. The figures are most useful for comparing models against each other on the same prompt under one consistent assumption. Do not rely on them for billing, budgeting, or financial reporting. For authoritative costs, consult your provider’s billing. pysolvr LLC makes no warranty that displayed cost figures match what any provider actually charges you, and accepts no liability for decisions made on the basis of estimated costs.

17. Responsibility for Prompts and Indemnity

You are solely responsible for every prompt, system instruction, conversation, attack string, and other input you submit through the Service, whether entered by you, generated by your software, or produced by an AI agent acting on your behalf; for the use to which you put the Service’s output, and for ensuring all of it complies with Section 3a (Acceptable Use), the applicable provider’s terms (Sections 10 and 11), and applicable law.

Provider terms apply to your prompts. Prompts you run are transmitted to third-party LLM providers and to any endpoint you register, each of which imposes its own acceptable-use and safety policies. A provider may rate-limit, suspend, or terminate access, or pursue remedies, in response to your usage; that is between you and the provider.

Indemnity. To the maximum extent permitted by law, you agree to indemnify, defend, and hold harmless pysolvr LLC and its LLM providers against any claim, loss, cost, damage, liability, account suspension, regulatory action, or expense (including reasonable legal fees) arising from or relating to: the prompts you run; your use of the Service as an attack vector or in any manner prohibited by Section 3a; your breach of any provider’s acceptable-use policy or terms (incorporated under Section 10); or any unlawful or unauthorised use of the Service by you, your software, or an agent acting on your behalf. This indemnity is not subject to any limitation of liability in these Terms. Because we access third-party providers on your behalf, a provider’s claim against us arising from your usage is your responsibility.

This section is in addition to, and does not limit, the general limitation of liability in Section 6, the programmatic-and-agentic-use terms in Section 15, and the acceptable-use terms in Section 3a.

18. Export Control and Sanctions

You represent and warrant that you are not located in, under the control of, or a national or resident of any country subject to comprehensive US sanctions or embargoes, and that you are not identified on any US government restricted-party or sanctions list. You will not use the Service, or permit it to be used, in violation of any applicable export control or economic sanctions laws, and you will not make the Service available to any person or entity barred under those laws. Our LLM providers impose equivalent restrictions that flow through to your use (Section 10).