Terms of use
Effective September 11, 2026. Trismag is operated by Jonathan Hill, an individual, not an LLC. Contact: Jonhillnj@gmail.com.
These terms apply to the Trismag website, hosted API, gateways and public preview. By using the service, you agree to these terms to the extent a valid agreement is formed under applicable law. If you do not agree, do not use the service. You must be an adult able to enter an agreement; when acting for an organization, you must have authority to do so. These terms do not retroactively remove existing rights or obligations.
1. An early warning, not a guarantee
Trismag flags known patterns associated with prompt injection in submitted text, conversations and tool metadata. It is a deterministic rule-based inspection service, without a language model in the detection loop. It evaluates content; it does not itself block an agent, control permissions, prevent execution, verify a tool’s behavior or guarantee protection from an attack.
We do not promise to detect, prevent or stop all prompt injection, data leakage, unauthorized actions, malicious tools or other threats. We make no claim of 100% detection, complete coverage, guaranteed safety, or a particular real-world detection rate. Existing, novel, indirect, multilingual, obfuscated and context-dependent attacks may pass undetected. Changes in threats or software can alter results.
LOW does not mean safe, trusted, approved or authorized. A higher risk level is a rule match, not proof of malicious intent. Legitimate content can trigger warnings. Scores and rule weights are not calibrated probabilities. An error, timeout, quota refusal, missing verdict or incomplete scan is not a successful inspection. Sanitized or review-formatted text may still contain harmful instructions and must not be treated as safe executable input.
Published evaluations describe their stated datasets, engine versions, thresholds and conditions. Development or regression results are not independent validation, real-world accuracy, a prediction of future performance or evidence that an agent was protected. Inspect both misses and benign flags in the evaluation.
2. Your integration and responsibilities
You decide whether and how to use a result. Keep independent security controls such as least privilege, restricted network access, tool isolation, output checks and explicit approvals for consequential actions. Define a failure policy and test it with your actual workflow. A LOW verdict must not be the sole basis for granting an agent permissions, transferring funds, disclosing sensitive data or taking another consequential action.
You are responsible for your agent, tools, credentials, integrations, input rights and actions. Submit only content you are authorized to process. Do not send secrets or regulated or highly sensitive records to the public demo. This preview is not designed to be the sole control in safety-critical systems, emergency services or decisions affecting health, liberty, employment, credit or other significant rights. Results are not professional, legal, financial, medical or compliance advice.
3. Preview access, limits and changes
The current public preview is free. Paid checkout and subscriptions are not available. Any internal pricing ideas are not an offer, promised future price or entitlement. Current quotas are published in the API’s health contract and may change. There is no promised uptime, throughput, response time, uninterrupted access, support response time or service-level agreement. Performance measurements are observations under their stated conditions, not guarantees.
We may update rules, interfaces or limits, suspend access for abuse or security reasons, and change or discontinue the preview. We will provide notice when reasonably practicable and when required by law, but urgent changes and outages may occur without advance notice. New paid offerings, if introduced, will require separately disclosed terms and affirmative purchase; using this preview does not authorize a charge.
4. Acceptable use and credentials
Do not use the service unlawfully, attack or disrupt infrastructure, evade quotas or authentication, obtain unauthorized information, impersonate another person, or use another person’s API key. Legitimate testing of suspicious content you are authorized to assess is part of the intended use. Production load tests or tests likely to disrupt service require prior coordination with the operator. Report vulnerabilities privately to the contact above and omit secrets from initial reports.
Protect your API key and report suspected compromise. You may stop using the service at any time and request account deletion as described in the privacy policy. Termination does not remove rights or obligations that by their nature continue, including applicable limitations, ownership and accrued claims.
5. Content, privacy and third parties
You retain rights you have in submitted content. You authorize the processing needed to provide the scan and return its result; these terms do not grant permission to sell your scans, publish them, or train models on them. The privacy policy describes content processing, retained metadata, providers, support email and deletion limitations.
Third-party hosts, browsers, agent frameworks and MCP servers are separate systems. Links, templates or an inspection do not endorse or certify them. You are responsible for their configuration and applicable terms. Hosted service access grants no ownership of Trismag’s software or branding. Any separately distributed source code is governed by its stated license; these terms do not override that license.
6. No warranties
To the maximum extent permitted by applicable law, the service and its results are provided “as is” and “as available.” Jonathan Hill disclaims express, implied and statutory warranties, including merchantability, fitness for a particular purpose, non-infringement, accuracy, completeness, security, availability and freedom from error. No example, score, demonstration, marketing statement or informal support exchange creates a guarantee of protection or performance.
Nothing here excludes a warranty, remedy or consumer protection that the law does not permit to be excluded. These terms are not permission to make misleading claims or disregard applicable privacy or security duties.
7. Limits of liability
To the maximum extent permitted by law, Jonathan Hill will not be liable for indirect, incidental, special, consequential, exemplary or punitive loss arising from use of or inability to use Trismag, including lost profits, business interruption, lost data or opportunities, or downstream actions taken by an agent, even if advised of the possibility.
To the maximum extent permitted by law, Jonathan Hill’s aggregate liability arising out of or relating to this free preview and these terms is limited to US $100. This limitation applies only where lawful. It does not exclude or limit fraud, willful misconduct, gross negligence, death or personal injury caused by negligence, or any other liability, statutory right or remedy that cannot legally be excluded or limited. Where a restriction is unenforceable, applicable law controls. A disclaimer does not eliminate every potential claim or liability.
8. Questions, disputes and updates
Please contact Jonhillnj@gmail.com with concerns so we can try to resolve them. This request does not impose a mandatory waiting period or remove access to courts, regulators, statutory deadlines or other legal remedies. These terms do not impose arbitration, a class-action waiver or an exclusive forum. Applicable law and mandatory jurisdiction rules determine rights and venue; no jurisdiction is invented or claimed here.
If a provision is unenforceable, the remaining provisions apply to the extent permitted by law. Failure to enforce a provision is not a waiver. Material updates will be posted with an effective date and additional notice or agreement obtained where required. If a future LLC takes over, we will identify the new operator and handle any transfer as required; these terms do not represent that transfer as already completed.