Terms of Service

These Terms are between Mantia Ltd ("Mantia", "we", "us") and the business that orders a Mantia service ("you"). They apply to every Mantia service: Mantia Essentials, which you buy online; managed assessments, which our team runs for you; and any service we add later. The details particular to each service are in its schedule at the end of these Terms.

Our services are for businesses only. You must be ordering in the course of your business, not as a consumer. If you accept these Terms on behalf of a business, you confirm that you are authorised to accept them for that business.

Key terms

Term	What it means
Us	Mantia Ltd, registered in England and Wales, company number 17172108, registered office 124-128 City Road, London, England, EC1V 2NX. Email: hello@mantia.co.uk
You	The business named in your order
Order	Your order for a service: for Mantia Essentials, your online order and its order summary; for a managed assessment, the order form we both sign; for any other service, as its schedule sets out
Service	The Mantia service you order, as section 2 and its schedule describe it
The agreement	These Terms (with these Key terms and the schedule for your service), the Data Processing Terms, and your order
Governing law	The law of England and Wales
Courts	The courts of England and Wales, exclusively
General Cap Amount	The greater of what you paid or must pay for the order the claim relates to and £250
Increased Claims	A breach by us of section 6 or the Data Processing Terms, or a breach by either of us of section 9
Increased Cap Amount	The greater of £5,000 and the General Cap Amount
Unlimited Claims	The claims listed in section 11.4
Data Processing Terms	https://mantia.co.uk/legal/data-processing
Notice address	Ours: hello@mantia.co.uk. Yours: the email for notices in your order form, or for Mantia Essentials, the email address you sign in with

1. How these Terms work

1.1 Services and schedules. These general terms apply to every service. The schedule for your service adds what is particular to it: Schedule 1 for Mantia Essentials and Schedule 2 for managed assessments. When we add a service, we add a schedule for it.

1.2 Which document wins. If we have signed a separate written agreement with you, including an order form for a managed assessment, that agreement wins over these Terms and the Data Processing Terms for anything it covers (for example, how long we keep your data, or opting out of lessons), and these Terms apply to everything else. But it cannot remove anything UK GDPR requires a processor contract to contain, and an order form cannot permit testing a live system (section 3.1(b)). Subject to that, the Data Processing Terms win on personal data (section 6.8), and the schedule for your service wins over these general terms for that service.

1.3 Accepting. For Mantia Essentials, you accept these Terms by ticking the box that says you agree to them at checkout and paying; no signature is needed. For a managed assessment, you accept them by signing an order form that refers to them; both of us sign it, in writing or electronically, including through an e-signature service. For any other service, you accept them as its schedule sets out. The schedule for your service says when the contract is made.

1.4 Language. These Terms and your order are in English only.

1.5 Industry codes. We have not signed up to any industry code of conduct for our services.

1.6 Copies of these Terms. These Terms and the Data Processing Terms are published at https://mantia.co.uk/legal/terms and https://mantia.co.uk/legal/data-processing, and every version of each stays published at https://mantia.co.uk/legal/terms/versions, where you can save or print it.

2. What our services are, and what they are not

2.1 What we do. We connect to your AI agent through the access you give us and run conversations built to break it: people trying to hijack it, people trying to make it say something you would regret, and questions that tempt it to make things up. An AI judge, made by a different company from the AI that plays the people in the conversations, rules on each conversation and quotes your agent's exact words. If the judge declines to rule on a conversation, a person at Mantia reviews it and rules instead, and your Report marks that ruling "Reviewed by the Mantia team". Our team reads the conversations as part of reviewing your results before we send them to you.

2.2 What we test. The tests that the schedule for your service describes.

2.3 What you get. Your results on a private, password-protected web page (the "Report"): every failure we found, with your agent's exact words and the turn where it happened; how serious each one is, and why; and every conversation in full. The schedule for your service may add to this.

2.4 What it is not. Each service is a point-in-time test of the version of your agent we can reach, with the evidence for everything we find. It is not:

(a) a certification or an audit, or an assurance that your agent is safe, secure, accurate or compliant with any law or standard;

(b) a promise that we will find every failure. If a test finds nothing, that shows only that these conversations did not make your agent fail;

(d) legal, regulatory or other professional advice.

2.5 Findings about your documentation. Some tests check your agent's answers against the documentation you give us. If your documentation is incomplete, a true answer can look unsupported, so these findings say "not supported by the documents you gave us", and we check each one before it goes out.

2.6 Our people and service providers. We decide how to run each service, using our own people, methods and tools. People who work for Mantia, the service providers listed on our sub-processors page and any other AI model provider we use under your order form help us provide it. We are responsible for what they do in providing it.

3. Your promises about the agent we test

3.1 You promise, from when you place your order until we have finished testing, that:

(a) you or your organisation control the agent's configuration and who can access it, and you are entitled to authorise us to test it in the way these Terms describe. Where your agent runs on another company's platform (for example Intercom, Ada or Copilot Studio), or uses another company's AI model, this includes that the terms of that platform and of that model provider allow this testing, or that you have their permission;

(b) what we test is a staging (test) version of your agent, not your live system; it holds no real customer data; and nothing the agent does there reaches real money, real messages to real people, or real staff. Testing a live system needs a separate written agreement, beyond an order form, that says so expressly;

(c) the key you give us was created only for Mantia, works only on that staging version, is not a live or personal key, and can be revoked by you at any time;

(d) you have the right to give us your documentation and everything else you give us, and none of it contains your customers' personal data; and

(e) what you tell us about your agent and how to connect to it is accurate.

3.2 Why this matters. We test only with the permission of whoever is entitled to control access to your agent, as the law requires (including the Computer Misuse Act 1990). We rely on your promises in section 3.1 for that permission.

3.3 If something changes. Tell us straight away if any promise in section 3.1 stops being true. We will pause testing, and section 13.1 then applies.

3.4 If a promise is untrue. You are responsible for the consequences of any promise in section 3.1 being untrue, and you will protect us against claims by others as section 12 sets out.

3.5 Helping us. You will give us what we reasonably ask for in order to connect and test, keep the staging version and the key working until you receive your results, and name a technical contact who answers by email. We are not responsible for being unable to provide the service because you did not do these things.

4. Testing is adversarial

4.1 What testing involves. Our tests are meant to make your agent fail. The people in our conversations, played by AI, may try to hijack it, talk it out of its rules, claim authority they do not have, ask it for dangerous information, push it towards hateful or abusive language, and play a customer who says they are thinking of harming themselves.

4.2 What your agent may say. Your agent may reply with offensive, false, harmful or disturbing content. The Report quotes your agent's exact words and includes every conversation in full, including the messages our side wrote, so it may contain that content. Share it within your organisation with that in mind.

4.3 Your systems. Our messages may set off your own monitoring, alerts, rate limits or handover to a person, and may be flagged by the platform or AI model provider your agent uses. Make sure nothing in your staging version passes our messages to real people or real systems.

4.4 Your costs. Anything it costs you to run your agent, your staging version and any platform while we test (for example model, hosting or per-conversation platform charges) is yours to pay.

4.5 Stopping us. The quickest way to stop testing is to revoke the key: that stops it at once, and counts as stopping it at your request. You can also email us at hello@mantia.co.uk, and we will stop as soon as we can after we receive your email. The schedule for your service says whether you get a refund.

5. Price, payment, tax and refunds

5.1 Price. The price is the one in your order summary or order form, in pounds sterling.

5.2 Payment. You pay as the schedule for your service sets out. Any currency conversion or card charges your bank makes are yours.

5.3 Tax. Prices do not include VAT. If we must charge VAT, your order summary or order form shows it. You are responsible for any other taxes, duties or levies that apply to your purchase where you are, but not for our income taxes.

5.4 Disputing a charge. If you think we have charged you wrongly, tell us within 30 days of the payment. We will work with you in good faith to resolve it within 15 days. If we cannot agree, either of us may use any remedy these Terms or the law allow.

5.5 Refunds in every service. The schedule for your service sets out its refunds. In every service you also get a full refund of what you paid for an order if: (a) we cannot put a shortfall in the Report right under section 10.3; (b) you end the order under section 13.2 before we deliver your Report; (c) we stop testing for a reason in section 13.1 other than a promise in section 3.1 being untrue or a breach of section 7; or (d) you cancel an order under section 2.6(a) of the Data Processing Terms after objecting to a new sub-processor or to someone new in our team working on your data. We refund to the card or account you paid from, or, if you have not yet paid, we cancel the charge or invoice.

6. Your data, our security, and who can see it

6.1 What we use your materials for. We use what you give us (your answers, documentation, your agent's instructions if you give them, connection details and key) and the conversations with your agent to provide the service to you. We may also draw lessons from your orders to improve Mantia, for example which of our techniques worked and how to sharpen our tests. We write those lessons in our own words and use them only to improve Mantia. They never include your data, your documentation, your agent's replies, your name, or anything else that could identify you or any person, and we never use your materials or the conversations to train an AI model (section 8.5). The Data Processing Terms cover any personal data involved. By accepting these Terms, you instruct us to anonymise any such personal data in order to draw those lessons, and the Data Processing Terms record that instruction. The lessons in this section apply to every service, unless your order form opts out of them.

6.2 Who can see it. Within Mantia, only the members of our team who work on your order can see your key and what you give us, and they read the conversations as part of the review. We tell you before anyone new in our team starts working on your data, as the Data Processing Terms set out.

6.3 Where it is held. The schedule for your service says where your data is held. Our team may access your data from outside the UK and the EU. That access is protected by: working only on our instructions; written confidentiality undertakings; the ICO's International Data Transfer Agreement where UK law requires one; disk-encrypted computers; and deletion on the dates in the schedule for your service.

6.4 AI model providers. Our AI model providers are Anthropic and OpenAI. The model that judges a conversation is always made by a different company from the model that plays the people in it. Our team also uses Anthropic's AI assistant, under Anthropic's commercial terms, to help review the conversations and prepare your Report. Our sub-processors page always shows the current list, what each provider does, where it processes data and how long it keeps it, with links to its own terms, and we tell you before we add or replace one, as the Data Processing Terms set out. We may change which provider does which part of the testing, and which of their models we use, for example when a newer one gives better results. For a managed assessment, the order form names the models we will use, and we change them only with your written agreement. If you ask for or agree to models from another provider, your order form says what that provider does, where it processes data, how long it keeps data and what protects data sent outside the UK; you approve it by signing, and we do not add it to our sub-processors page.

For the testing, we send our AI model providers your documentation, our messages and your agent's replies. Your agent's instructions, if you give them to us, go only to the AI that prepares the tests, to the judge, and to the AI assistant when our team reviews your results. We never give them to the AI that plays the people in the conversations; if your agent gives them away in a reply, that is a failure we report. We never paste your key into the AI assistant. Under the terms we use them on, no provider uses what we send to train its models. Each keeps and deletes what we send under its own terms. Each may keep it longer where the law requires, or to enforce its usage policies or protect against harm, and because our conversations are adversarial, some may be flagged and kept longer for that reason. We cannot delete their copies sooner than their own terms allow. What our team puts into the AI assistant stays in our workspace until we delete it on the dates in the schedule for your service.

6.5 How we protect it. For every service: within Mantia, only the members of our team who work on your order can reach your data, each with their own logins, and every account of ours that holds your data uses two-factor sign-in; the computers our team uses are disk-encrypted; on the computer that runs your testing, your key is kept in a private file; the only connection code in our code repository is generic, with no customer's details in it, and nothing specific to you (such as the connection code we write for your agent, your agent's address or other identifiers, your key or your documents) ever goes into it; we never paste your key into the AI assistant; the results are kept in our file storage with Google in a disk image we encrypt ourselves (AES-256); and your Report is protected by Netlify's password protection. The schedule for your service may add to this.

6.6 How long we keep it. We delete your data on the dates in the schedule for your service. "Deleted" means deleted from everywhere we hold it for that service: our order site's storage, our results site, the computers our team uses, our file storage with Google (including its bin), our mailbox, our Notion workspace and our AI assistant workspace. Our providers then remove deleted data from their own systems on their own schedules, and our AI model providers keep and delete their copies as section 6.4 describes. For each order that goes ahead (one you pay for, or an order form we both sign), we also keep: your order record (who ordered, what, when and for how much, and its status and times), the record of your acceptance of these Terms with the versions of these Terms and the Data Processing Terms you accepted, and payment records, for 6 years after the end of the financial year of your order; and our emails with you for 2 years after the last message, except any email that carries your documentation, the conversations or your findings, which we delete with your documentation. None of these records includes your key or your documentation. Our order folders and the private file that holds your key are never included in backups of our computers.

6.7 Data about how the service ran. We may record data about how the service ran for your order, such as timings, costs, the number of conversations and the number of findings for each test, and use it to run and improve Mantia. It never includes the content of your documentation or the conversations. We share it only combined with other customers' data, so that it does not identify you.

6.8 Personal data. Our Data Processing Terms (https://mantia.co.uk/legal/data-processing) apply to any personal data in what you give us to test your agent and in the conversations, which we handle for you. If they conflict with these Terms about personal data, the Data Processing Terms apply. Our Privacy Policy (https://mantia.co.uk/legal/privacy) explains how we handle the personal data that is ours: your people's contact details and your order, sign-in and payment details. For any question or request about your data or personal data, email hello@mantia.co.uk.

6.9 What you must not send. Do not give us real customers' personal data, payment card or bank account numbers, government identity numbers, or special category data (such as health information).

7. Acceptable use

You will not, and will not let anyone else:

(a) ask us to test an agent or environment that you are not entitled to authorise us to test, or a live system;

(b) use a service, the Report or the conversations to attack, or to gain unauthorised access to, anyone's systems;

(d) except as section 8.3 or the law allows, copy, reverse engineer or try to discover our methods, prompts, tests or software, or use a service, the Report or the conversations to build a competing service;

(e) run security tests on, interfere with, or get around the access controls of our order site, results pages or other systems;

(f) let anyone use your sign-in, or give your Report's password to anyone other than the people section 8.3 allows you to show your Report to; or

(g) use a service for anything unlawful.

8. Who owns what

8.1 Your materials. You keep all your rights in what you give us and in your agent's replies. You allow us to copy, use and change them only as needed to provide the service, to draw the lessons in section 6.1 and to record the data in section 6.7. You are responsible for their accuracy and content.

8.2 Your Report. When we deliver your Report, we assign to you our rights in it, including the findings about your agent, except for Our Materials (section 8.3). If you ask, we will confirm that assignment in a document signed by us.

8.3 Our Materials. We keep everything we owned before your order or develop outside it; our tests, goals, prompts, the messages our side writes in the conversations, our judge, our severity scale, the layout of the Report and our software, including any we create during your order; and the lessons we draw under section 6.1 (together, "Our Materials"). Where Our Materials appear in your Report or the conversations, you may use them, for as long as you like, to understand and fix your agent, to test it again yourself, and to show what we found to people in your organisation, your advisers, auditors, insurers and regulators, and organisations that deploy your agent or assess it for their own use, under written confidentiality terms. You may not give Our Materials to a company that sells AI testing, publish them or sell them.

8.4 AI-written content. The messages our side writes, and our judge's rulings and explanations, are written by our model providers' AI models, whose terms may limit using that output to train other AI models. So you will not use them to train or fine-tune an AI model. Your agent's own replies are yours to use as you like. You may use your Report, our findings and your agent's replies to understand and fix your agent, and your agent's own replies to retrain it.

8.5 No training on your data. We do not use what you give us, the conversations or your Report to train any AI model.

8.6 Feedback. You do not have to give us feedback. If you do, you give it as it is, and we may use it freely.

8.7 Nothing else transfers. Apart from what this section says, neither of us transfers any rights in its products, data or other intellectual property to the other.

9. Confidentiality

9.1 Keeping it confidential. Each of us will use the other's Confidential Information only to provide or receive the service under these Terms (and, in our case, to draw the lessons in section 6.1 and to record the data in section 6.7), and will not disclose it to anyone else except as this section allows. Each of us will protect the other's Confidential Information at least as carefully as its own similar information, and never with less than reasonable care.

9.2 What it covers. Confidential Information is information one of us gives the other in connection with these Terms that is marked confidential or should reasonably be understood to be confidential. Yours includes what you give us, your key, the conversations, your Report and the findings. Ours includes Our Materials and anything non-public about how our services work.

9.3 What it does not cover. Information the recipient already knew without a duty of confidence, that is or becomes public through no fault of the recipient, that the recipient receives without a duty of confidence from someone entitled to give it, or that the recipient develops independently without using the other's Confidential Information.

9.4 Disclosures the law requires. A recipient may disclose Confidential Information as far as the law requires, if (where the law allows) it gives the other reasonable notice first and reasonably helps, at the other's cost, any effort to keep it confidential.

9.5 Who else may see it. Each of us may disclose the other's Confidential Information to its staff, contractors, advisers and representatives who need to know it, if they are bound to keep it confidential (by contract, professional duty or law) at least as strictly as this section requires; each of us is responsible for their compliance. We may also share yours with the service providers on our sub-processors page and any other AI model provider we use under your order form, only as needed to provide the service and under their own terms, as section 6 describes.

9.6 Your findings stay private. Except as sections 9.4 and 9.5 allow, we never publish or share your Report or your findings without your written agreement, except as combined data under section 6.7.

9.7 Naming you. We will not announce publicly that you are a customer without your written agreement. We may mention privately, to investors and prospective customers, that you are a customer, unless you tell us not to.

9.8 When an order ends. We delete your Confidential Information as section 6.6 and the schedule for your service set out. You may keep your Report and the conversations, subject to section 8.

10. Our promises, and what we do not promise

10.1 Both of us. Each of us promises the other that it has the power to accept these Terms, is properly established under the laws of the place where it was formed, and will comply with the laws that apply to it in carrying them out.

10.2 Us. We promise that: (a) we will perform the service with reasonable care and skill, in a competent and professional way, and as the schedule for your service describes; (b) the Report, apart from your materials and anything written by AI, will not infringe anyone's copyright, trade mark or trade secret; and (c) the Report will contain what section 2.3 describes.

10.3 If the Report falls short. If the Report does not contain what section 2.3 describes, tell us within 14 days of delivery, with enough detail for us to see the problem. Within 10 business days of your message we will put it right at no cost, by correcting the Report or, if you give us a new key, running the affected tests again. If we cannot, section 5.5(a) applies. Apart from that refund, this is your only remedy for that problem.

10.4 AI can be wrong. Our side of the conversations is written, and most verdicts are reached, by AI models, and AI can be wrong. Our team reads the conversations as part of the review and rules itself where the judge declines (section 2.1), but we do not promise that every verdict is right or that we have found every failure. AI-written text may resemble text written for others and may not be protected as intellectual property.

10.5 Nothing else. Apart from the promises in these Terms, all other terms, conditions and warranties that might be implied by law, statute or custom are excluded, as far as the law allows.

11. Limits on liability

11.1 What we never limit. Nothing in these Terms limits or excludes either of our liability for death or personal injury caused by negligence, for fraud or fraudulent misrepresentation, or for anything else the law does not allow to be limited or excluded.

11.2 Caps. Subject to sections 11.1 and 11.4, the total liability of each of us arising out of or in connection with an order, whether in contract, tort (including negligence), breach of statutory duty or otherwise, is no more than: (a) the Increased Cap Amount, for Increased Claims; and (b) the General Cap Amount, for all other claims.

11.3 Losses neither of us is liable for. Neither of us is liable to the other for loss of profits, revenue, business, contracts, goodwill or anticipated savings, or for any indirect or consequential loss, however it arises. This section does not apply to Increased Claims, a breach of section 9, or your obligations under section 12.

11.4 Unlimited Claims. The caps in section 11.2 do not apply to these "Unlimited Claims": your obligation to pay; your obligations under section 12; your breach of section 3.1, 6.9 or 7; anything in section 11.1; and any liability that the Data Processing Terms say these Terms do not limit.

11.5 Why these limits. Each order is a test of fixed scope at a price agreed in advance. You control the agent we test and decide what to give us, and the key you create opens only a staging version with no real data. If you need different terms, ask us for a separate written agreement.

12. Your protection of us

12.1 What you cover. You will indemnify us against all claims made against us by anyone else, and all resulting damages, awards, settlements, costs and expenses (including reasonable legal fees), arising from: (a) a promise in section 3.1 being untrue; (b) our use of what you give us, as these Terms allow, infringing anyone's rights; (c) your breach of section 7; or (d) your fraud or wilful misconduct.

12.2 How it works. We will tell you promptly about any claim, give you reasonable help at your cost, and let you control its defence and settlement. We may take part with our own lawyers at our own cost. You will not settle a claim in a way that admits fault on our part or otherwise materially harms us without our written agreement.

12.3 Our other rights. This section is our only remedy for a claim it covers. It does not limit our right to recover our own losses from a breach of these Terms.

13. Stopping and ending an order

13.1 When we may stop. We may pause or stop testing at once, and tell you why, if we reasonably believe that a promise in section 3.1 is untrue, that testing could affect a live system, real people or someone else's system, or that continuing would break the law or our model providers' rules. Section 5.5 and the schedule for your service say whether you get a refund.

13.2 Ending for breach. Either of us may end an order at once by notice, giving the reason, if the other materially breaches these Terms and, where the breach can be put right, has not put it right within 5 business days of being told about it.

13.3 Events beyond our control. Neither of us is liable for a delay or failure caused by an unforeseen event outside its reasonable control (such as a natural disaster, war, pandemic, riot, act of terrorism, or a failure of the internet or a utility), if it took reasonable steps to avoid or reduce the effects. The refunds in section 5.5 and the schedule for your service still apply.

13.4 What continues. Sections 2.4, 3.4, 4.2, 5, 6, 7, 8, 9, 10, 11, 12, 13.4, 14 and 15, and the refund and data terms of the schedule for your service, continue after an order ends, as do any other terms that by their nature should.

14. General

14.1 Whole agreement. The agreement described in the Key terms, with any separate written agreement under section 1.2, is the whole agreement between us about your order, and replaces anything said or written about it before. Nothing in this section limits liability for fraud. Terms in your purchase order or similar documents do not apply.

14.2 Changes. We may change these Terms, including by adding a schedule for a new service, for future orders by publishing a new version. The version you accepted applies to your order. Any other change must be agreed in writing by both of us.

14.3 If part is invalid, and waiver. If a court finds any part of these Terms invalid or unenforceable, the rest stays in force. Not enforcing a term, or not exercising a right, is not a waiver of it.

14.4 Law and courts. These Terms, and any dispute or claim arising out of or in connection with them (including non-contractual disputes or claims), are governed by the law of England and Wales. The courts of England and Wales have exclusive jurisdiction.

14.5 Urgent relief. Despite section 14.4, a breach of section 9 or of either of our intellectual property rights may cause harm that money cannot put right, so the one harmed may seek an injunction or other urgent relief in any court that has jurisdiction.

14.6 Remedies. Except where these Terms say a remedy is the only one, using one remedy does not limit any other.

14.7 Transfer. Neither of us may transfer its rights or obligations under these Terms without the other's written agreement, except to a successor on a merger, change of control, reorganisation or sale of all or substantially all of the business these Terms relate to, after telling the other. Any other transfer is void.

14.8 Notices. Formal notices under these Terms must be in writing and sent by email to the notice address in the Key terms. A notice is received when sent, unless the sender gets a message that it was not delivered; if it is sent outside 9:00 to 17:00 UK time on a business day, it is received at 9:00 UK time on the next business day. Emails about running your order (for example our email saying testing has started, or yours cancelling before it) take effect when sent.

14.9 Independent parties. We are independent contractors, not agents, partners or joint venturers. Neither of us can bind the other.

14.10 No third-party rights. No one else has any rights under these Terms. The Contracts (Rights of Third Parties) Act 1999 does not apply.

14.11 Sanctions and export controls. You will not use a service, or allow it to be used, in breach of any sanctions or export control law that applies to you or to us, including those of the United Kingdom and the United States. You confirm you are not a person whom those laws prohibit us from dealing with.

14.12 Anti-bribery. Neither of us will do anything that would break a law against offering, giving, promising, or receiving, directly or indirectly, money or anything of value to win or keep business, such as the UK Bribery Act 2010.

14.13 Interpretation. Headings are for convenience only. "Including" and similar words do not limit what comes before them.

14.14 Accepting online. Where you accept these Terms by ticking a box, that acceptance has the same effect as a signature.

15. Definitions

Business day: Monday to Friday, except bank holidays in England. UK time: the time in London. Confidential Information: section 9.2. Data Processing Terms: the terms at the link in the Key terms. General Cap Amount, Increased Claims, Increased Cap Amount: the Key terms. key: the key, token, test account or other access you give us to reach your agent. Order: the Key terms. Our Materials: section 8.3. Report: section 2.3. Service: the Key terms. Sub-processors page: our published list of the service providers that process data for us (apart from any other AI model provider we use under a managed assessment's order form, section 6.4), at https://mantia.co.uk/legal/sub-processors. The conversations: the conversations between our AI and your agent during testing, including the messages on both sides. Unlimited Claims: section 11.4.

Schedule 1: Mantia Essentials

S1.1 What it is. Mantia Essentials is a fixed-price test of your AI agent that you order and pay for online at app.mantia.co.uk.

Detail	For Mantia Essentials
Tests	Those listed on the Mantia Essentials product page when you place your order (S1.4)
Price	The price in your order summary, paid in full when you order (S1.7)
Results	Target: within 24 hours of testing starting (S1.5)
Refunds	Full refund if we can't connect to your agent, if you cancel before testing starts, if your results are more than one business day late, or if you don't send what we need within 5 business days of our request; also in the other cases in S1.6 and section 5.5

S1.2 Who it is for. AI agents that meet the requirements on the Mantia Essentials product page when you place your order: in short, agents that talk to people in text, through an API that we can reach from the internet.

S1.3 How your order is made.

(a) The steps. You order on app.mantia.co.uk: you sign in with your work email and a code we send you; tell us about your agent and give us your documentation and connection details, including a key made just for us; tick the boxes that confirm your authorisation to have the agent tested and accept these Terms; then check the order summary and pay through Stripe.

(b) Correcting mistakes. Your answers save as you go. Until you pay, you can go back to any step and change what you entered. The order summary before payment shows the agent, the company and the price.

(c) When the contract is made. Your order is your offer to buy. The contract is made when we send the order confirmation email, which we send as soon as Stripe confirms your payment. Our emails about your order come from essentials@mantia.co.uk. Once you tick the box accepting these Terms, sections 3, 6, 7, 9, 11, 12, 14 and 15 and the Data Processing Terms apply, including to anything you gave us before then, even if you do not go on to pay. If we do not accept your order, we will tell you and refund in full what you paid.

(d) Keeping a copy. We keep a record of your order and of the versions of these Terms and the Data Processing Terms you accepted, with the time you accepted them (section 6.6). Every version stays published (section 1.6), and you can ask us for a copy of your order and the versions you accepted at any time.

(e) Your sign-in. Keep access to your work email secure: anyone who can read it can sign in as you. Tell us at once if you think someone else has signed in to your order.

S1.4 What we test. The tests listed on our product page for Mantia Essentials, at https://mantia.co.uk/essentials, when you place your order. We keep a dated copy of that page for each version of it, and will send you the copy that applied to your order if you ask.

S1.5 Results within 24 hours: what that means. Where we say results within 24 hours, we mean within 24 hours of a successful connection to your agent and the start of testing. This section sets out in full what that means. We send one message at a time, never several in parallel.

(a) Connecting. After you pay, we build the connection to your agent from what you gave us. A successful connection means your agent has answered our check messages through that connection, carrying a conversation on from one message to the next. If we need anything else from you, we email you and wait for your reply.

(b) Testing starts. Once we have a successful connection and your documentation is ready for the tests, we start testing and email you to say so, with the time your results are due: 24 hours after that email. Testing starts when we send that email. We send it only on a Monday, Tuesday, Wednesday or Thursday, UK time, so your results are due on a weekday.

(c) No promise about when testing starts. We do not promise when testing will start. If we cannot make a successful connection within 2 business days of having everything we asked you for, we will refund you in full (S1.6).

(d) Delivery. Your results are delivered when we send the email with the link to your Report. The password follows in a separate email.

(e) A target, with a refund. 24 hours is a target, not a guarantee. If we see that we will miss the due time, we will email you before it passes, with a new time. If your results arrive more than one business day after the due time, you get a full refund and still receive your results.

(f) If your agent stops answering. If, after testing has started, your agent or staging version stops answering, or the key stops working, for a reason on your side other than your revoking it, we will pause testing and tell you. The due time moves back by as long as testing was paused. If testing is still paused 5 business days after we tell you, we will close your order and refund you in full.

S1.6 Refunds.

(a) The rule. Full refund if we can't connect to your agent, if you cancel before testing starts, or if your results are more than one business day late.

(b) What each part means. We can't connect: we have not made a successful connection within 2 business days of having everything we asked you for, or we stop for any reason before testing starts. You cancel before testing starts: you email hello@mantia.co.uk before we send the email saying testing has started. Your results are more than one business day late: as S1.5(e) describes.

(c) Other full refunds. You also get a full refund in the cases in section 5.5, if we close your order under S1.5(f), and if we ask you for anything we need in order to connect or to start testing and you do not send it within 5 business days of our request (we then treat your order as cancelled before testing started).

(d) How you are refunded. We refund everything you paid to the card you paid with, through Stripe. You do not need to fill in a form. Refunds usually appear within 5 to 10 days, depending on your bank.

(e) When there is no refund. There is no refund if testing stops at your request after it has started (section 4.5), or if we stop testing after it has started because a promise in section 3.1 was untrue or you breached section 7.

(f) The only remedy for these events. A refund for our being unable to connect, or for late results, is your only remedy for that event.

S1.7 Payment. You pay the full price when you order, by card through Stripe. A promotion code may reduce the price.

S1.8 Where your data is held. Your order, the documents you upload and your key are stored by our hosting provider, Netlify, in Frankfurt (EU), and cached on Netlify's global network when read. Our order site's server code runs in Frankfurt too. Netlify serves our sites, including your Report, from its global network, and keeps server logs as its own terms describe. Testing runs on the computer of whichever of our team prepares your order: it downloads your key into a private file and your documentation into your order's folder, and keeps them until the dates in S1.10. The results of the testing are kept in our file storage with Google in an encrypted disk image, and are also held on the computer of whichever of our team reviews them. Your Report is a website on Netlify.

S1.9 Extra protection for Mantia Essentials. Your order is kept in Netlify's standard encrypted storage, private to our order site, and each of our team reaches it with their own Netlify login using two-factor sign-in and their own access token. We do not add our own encryption to what Netlify stores.

S1.10 How long we keep it.

(a) Your key: deleted 7 days after we deliver your Report. Please also revoke it yourself once you have your results.

(b) Your order data (apart from the records in section 6.6), your documentation, the connection code we write for your agent, the results in our file storage with Google, our team's AI assistant chats and project files about your order, our other copies of the conversations, our working files for your order, and any email that carries your documentation, the conversations or your findings: deleted 30 days after we deliver your Report.

(c) Your Report, including the conversations shown in it: taken down and deleted 90 days after delivery, unless you ask us at hello@mantia.co.uk to delete it sooner or to keep it for up to 12 months after delivery. Save anything you want to keep before it is taken down.

(d) If you do not pay, what you gave us is deleted 7 days after you last saved it.

(e) If your order ends without a Report (for example because we refunded it), everything in (a) and (b) is deleted 7 days after it ends.

(f) Your sign-in details: deleted 2 years after you last sign in.

Schedule 2: Managed assessments

S2.1 What it is. A managed assessment is a test of your AI agent, designed for you and run by our team. The scope, dates, price, payment and other details are in an order form that we both sign. The order form wins over these Terms for anything it covers, as section 1.2 sets out. This schedule sets the defaults that apply where the order form is silent.

S2.2 When the contract is made. The contract is made when both of us have signed the order form, in writing or electronically, including through an e-signature service. Sections 3, 6, 7, 9, 11, 12, 14 and 15 and the Data Processing Terms apply from when you sign it, including to anything you gave us before then. As we are both businesses, we agree that regulations 9(1) and 11(1) of the Electronic Commerce (EC Directive) Regulations 2002 do not apply to a managed assessment.

S2.3 Your authorisation. The order form includes your written authorisation for us to test your agent, signed for you by someone entitled to give it, and confirms your promises in section 3.1.

S2.4 What we test. The scope set out in the order form. We write the rules for your assessment from your own documents (your "goal sets"). The structure and method of the goal sets are Our Materials (section 8.3); the rules in them that come from your documents stay your Confidential Information, and we delete them as S2.9 sets out.

S2.5 Payment. You pay as your order form sets out. We invoice only once both of us have signed it.

S2.6 What you get. At least the Report described in section 2.3, on a password-protected results site, by the date in the order form. The order form may add more.

S2.7 Cancelling and refunds. Unless the order form says otherwise: you get a full refund if you cancel by emailing hello@mantia.co.uk before testing starts, and no refund if you cancel after that; if you have not yet paid, the price stays payable. Testing starts when we email you to say it has started. If we stop before testing starts, for any reason, you get a full refund and nothing more is payable. You also get a full refund in the cases in section 5.5. The order form may set a remedy if your Report is late; this schedule sets none.

S2.8 Your materials and where they are held. You send us, and we receive, your documents, key and other materials as agreed in your order form. Testing runs on the computer of whichever of our team runs your assessment, which holds your key in a private file and your documentation in your assessment's folder. Nothing specific to you goes into our code repository (section 6.5). The results are kept in our file storage with Google in an encrypted disk image, and your Report is a website on Netlify. Your goal sets are kept in our Notion workspace for the length of the engagement.

S2.9 How long we keep it. We keep your materials for the periods in your order form. Where it sets none, these apply:

(a) your key: deleted 7 days after we deliver your Report;

(b) your documentation, the connection code we write for your agent, the results in our file storage with Google, our team's AI assistant chats and project files about your assessment, our other copies of the conversations, our working files, and any email that carries your documentation, the conversations or your findings: deleted 30 days after we deliver your Report;

(c) your Report, including the conversations shown in it: taken down and deleted 90 days after delivery, unless you ask us at hello@mantia.co.uk to delete it sooner or to keep it for up to 12 months after delivery;

(d) if the engagement ends without a Report, everything in (a) and (b): deleted 7 days after it ends; and

(e) your goal sets: deleted from our Notion workspace, including its trash, on the end date of the engagement in the order form or, if it sets none, 30 days after we deliver your Report.

S2.10 Documents sent before an order form is signed. If you send us documents before an order form is signed, and none is signed, we delete them within 30 days after we receive them, unless a non-disclosure agreement signed by both of us covers them, in which case it governs them.

These Terms are adapted by Mantia Ltd from three standard agreements by Common Paper, each licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/): the Professional Services Agreement Standard Terms, version 1.1 (https://commonpaper.com/standards/professional-services-agreement/1.1); the Cloud Service Agreement Standard Terms, version 2.1, with its Terms of Service cover page (https://commonpaper.com/standards/cloud-service-agreement/2.1 and https://commonpaper.com/standards/terms-of-service/); and the AI Addendum, version 1.0 (https://commonpaper.com/standards/ai-addendum/1.0). Mantia has changed and combined them. Common Paper has not reviewed or endorsed these Terms.