1. Why this Policy exists
1.1 detumble.ai LLC ("detumble", "we") runs a shared platform. One customer's misuse can harm another customer's data, our providers' trust in us, and the people who live and work in the buildings our customers design. This Acceptable Use Policy ("Policy") sets the rules. It forms part of the Terms of Service; capitalised terms have the meanings given there.
1.2 The rules apply to you, to every User on your account, and to anything an integration of yours does through the API or MCP endpoints.
2. Comply with the law
2.1 You may use the Service only for lawful purposes and in compliance with all laws that apply to you and to the project you are working on, including intellectual property, confidentiality, data protection, building and planning, procurement, anti-bribery, sanctions and export control laws.
2.2 You may not use the Service to upload, store, process or produce anything that is unlawful in Egypt, in the country where you are, or in the country where the project is.
3. Upload only what you are entitled to upload
3.1 Rights in drawings. Construction drawings are usually the copyright of the architect, engineer or client who commissioned them, and are often confidential under a tender or contract. You may upload a drawing set only if you own it or have permission (a licence, a contract, a tender invitation, an employer's authority) to process it with a third-party service. You are responsible for checking this before you upload.
3.2 Confidential and classified projects. Do not upload drawings of facilities that are classified, security-sensitive or subject to a non-disclosure regime that forbids cloud processing (for example military sites, critical infrastructure control rooms, certain government buildings) unless the owner has authorised processing by us and our subprocessors in the locations disclosed in the Privacy Policy.
3.3 Personal data. Drawing sets sometimes contain names, signatures, contact details or, occasionally, more sensitive information. Remove what the project does not need. Do not upload special-category personal data (health, biometric, criminal, religious, political and similar) or data about children.
3.4 Malicious files. Do not upload files that contain malware, scripts intended to exploit our parsers, zip bombs, or content designed to break the reading layer or the engine.
4. Do not attack the Service, its providers or other customers
You may not, and may not attempt to:
(a) access any account, Project, wallet, schema, storage prefix or graph namespace that is not yours, or test whether you can;
(b) probe, scan or test the vulnerability of the Service, or breach or circumvent any authentication, authorisation, isolation, rate-limiting or security control, without our prior written permission;
(c) interfere with the operation of the Service, including by denial-of-service, flooding the API, creating an unreasonable number of accounts, Projects or Runs, or generating load that degrades the Service for others;
(d) introduce malware, or use the Service to distribute it;
(e) forge headers, spoof identities, impersonate detumble, its founders, its staff or another customer, or misrepresent your affiliation with any person or organisation;
(f) use the Service to attack, scan, spam or defraud any third party;
(g) use the Service to send unsolicited bulk messages; or
(h) harass, threaten or abuse our staff or other customers, including in chat messages that are reviewed for support or abuse.
5. Do not defeat the caps, gates, isolation or metering
5.1 The Service protects your money and ours with quotes, reservations, per-run budgets, account caps, a company-wide daily cap and a kill switch, and protects your professional responsibility with Review Gates and evidence pointers. You may not attempt to bypass, disable, spoof, replay or interfere with any of these, including by manipulating quotes, replaying webhooks, exploiting race conditions in the ledger, scripting acceptance at Review Gates, or removing evidence pointers, version stamps or disclaimers from Outputs.
5.2 You may not share sign-in links or sessions, run more than one person on a personal account, or use the API to give access to the Service to people who are not your Users.
6. Do not extract or copy the engine
You may not, and may not attempt to:
(a) reverse engineer, decompile, disassemble or otherwise derive the source code, architecture, algorithms, rule files, packs, ontologies, estimators or system instructions of the Service;
(b) prompt, trick or otherwise induce the Service or its models to reveal their prompts, system instructions, tools, hidden context or internal reasoning, or to act outside the tools we provide (prompt injection, jailbreaking);
(c) use the Service, or a systematic collection of Outputs, to build a data set for training a machine learning model, or to build, benchmark or improve a competing product;
(d) scrape, crawl or bulk-download the Service other than exporting your own Customer Content and Outputs through the interfaces we provide; or
(e) resell, rent, sublicense or time-share access to the Service.
6.2 Liability. An attempt under (a) or (b), whether or not it succeeds and however it is framed (a question, a role-play, an "audit", an encoded or split request, text placed inside an uploaded drawing, document or review answer), is a material breach that cannot be cured: the account is terminated without refund, you are liable for our loss and our costs of investigation and enforcement, you indemnify us for any use of what was obtained, and the limitation of liability in the Terms does not apply (Terms section 5.9). Section 12.6 describes how we detect and record such attempts.
7. Uses our AI provider prohibits
7.1 The Service uses large language models provided by Anthropic under Anthropic's Usage Policy, which we are contractually required to flow down. In addition to the rules above, you may not use the Service, its chat, or its models to:
(a) plan, develop or facilitate weapons, explosives, or chemical, biological, radiological or nuclear materials, or facilities whose primary purpose is to produce them;
(b) plan or facilitate attacks on critical infrastructure, or produce content intended to gain unauthorised access to any system;
(c) create malware, exploits or denial-of-service tools;
(d) produce content that incites violence, terrorism or hatred against protected groups, or that harasses, threatens or sexualises anyone, or any content involving the sexual abuse of children;
(e) deceive people about who or what they are talking to, generate disinformation, or interfere with elections or democratic processes;
(f) profile or surveil individuals, make decisions about individuals in criminal justice, employment, housing, credit, insurance or similar contexts, or process personal data without a lawful basis;
(g) commit fraud, counterfeiting, phishing, or deceptive commercial practices; or
(h) circumvent Anthropic's or our safety controls.
7.2 Anthropic's Usage Policy requires that in high-risk domains a qualified professional review AI-assisted output before it is relied on, and that people are told when AI assisted a decision. The Service's Review Gates and the AI Outputs and Responsibility Disclosure implement that requirement for construction. You must not remove or bypass them and must tell your own clients that AI assisted the work where the law or your professional rules require it.
8. Use Outputs responsibly
8.1 Outputs are computed estimates for professional review. You may not:
(a) present an Output to a client, authority, contractor, supplier or insurer as a verified, certified, sealed or stamped deliverable, or as the work product of a licensed professional, unless a licensed professional has actually reviewed and adopted it;
(b) use an Output as the sole basis for a structural, safety-critical or life-safety decision without professional verification;
(c) remove or alter the disclaimers, evidence pointers, version stamps or gate status that the Service places on an Output;
(d) represent detumble as the engineer, architect, surveyor or quantity surveyor of record on any project; or
(e) use the Service for a project where you know that no qualified professional will review the Outputs.
8.2 The AI Outputs and Responsibility Disclosure explains why these rules exist and what the Service does to help you meet them.
9. Use of the API and integrations
9.1 API and MCP access is available on plans that include it. You must keep API keys confidential, rotate them if they may have leaked, respect documented rate limits and caps, identify your integration truthfully, and comply with this Policy in everything your integration does. You may not use the API to give third parties access to the Service, to build a competing service, or to circumvent metering.
9.2 We may throttle, suspend or revoke API access that harms the Service or other customers.
10. Security research
10.1 We welcome good-faith security research. If you find a vulnerability, report it to support@detumble.ai with the subject line "Security" and give us a reasonable time to fix it before any public disclosure. Do not access, modify or exfiltrate other customers' data, do not degrade the Service, and do not use social engineering, physical attacks or denial-of-service.
10.2 We will not pursue legal action against researchers who follow Section 10.1, and we will treat the report as authorised access under this Policy.
11. Reporting abuse
11.1 If you believe someone is using the Service in breach of this Policy, or that Customer Content on the Service infringes your rights, e-mail support@detumble.ai with the subject line "Abuse" and give us enough information to identify the content or conduct and to contact you.
11.2 For copyright complaints, include the work you say is infringed, where on the Service the infringing material is, a statement that you have a good-faith belief the use is not authorised, a statement that your notice is accurate and that you are the rights-holder or authorised to act for them, and your contact details. We will pass the complaint to the account holder, who may respond.
12. Enforcement
12.1 We investigate reports and our own signals (security alerts, cap triggers, provider abuse flags). We may look at the relevant Customer Content, Outputs, chat messages and logs to do so; such access is limited to what the investigation needs and is logged.
12.2 Depending on the severity, we may: warn you; require you to remove content or change a behaviour; remove or disable content; suspend a Run, a Project, a User, API access or the account; terminate the account; report unlawful conduct to authorities; or take legal action. We aim to be proportionate: a first, minor, accidental breach gets a warning; deliberate attacks, fraud, exploitation of children, or threats to safety get immediate termination.
12.3 We will tell you what we did and why, unless the law prohibits it or telling you would endanger an investigation or another person.
12.4 Appeals. If you believe we acted in error, reply to our notice or write to support@detumble.ai with the subject line "Appeal" within 30 days. A founder who was not involved in the original decision will review it and reply within 10 business days.
12.5 Credits on Runs interrupted by an enforcement action are released. Refunds after termination are governed by the Fees, Credits and Refunds Policy.
12.6 Automated review of prompts and outputs. Every message, review answer and instruction you send to the Service, and every text the Service writes back to you, is reviewed automatically for attempts to extract Engine Internals (section 6). A request that is judged to be such an attempt is refused before it reaches a model, an output that would reveal Engine Internals is withheld or redacted, and the attempt is recorded (account, Project, time, the kind of attempt, a short excerpt) and alerts our security team. Repeated attempts freeze the account for chat and new Runs until a founder reviews it. These records are kept for the period stated in the Privacy Policy and may be used as evidence of a breach. The review is narrow: it looks for extraction and injection patterns, not for the content of your project, and an ordinary engineering question that happens to use words like "script", "model" or "prompt" is allowed.
13. Changes
We may update this Policy under the Terms of Service. Material changes are announced 30 days in advance, except where a change is needed urgently for security or to comply with a provider's policy or the law.
14. Contact
Abuse and security reports, and general questions: support@detumble.ai (subject "Abuse" or "Security")
Parts of this policy are adapted from the use restrictions of the 37signals Terms of Service by 37signals, licensed under CC BY 4.0, and changed for detumble.ai.