Your data. Your rules.

Yes, when it is set up with clear limits and inside accounts you control. The risk with AI is rarely the tool itself - it is an assistant given more access than it needs, or rules nobody wrote down. We set up every assistant inside your own accounts, you choose what it may send, pay or delete, and you get those rules in writing. The National Cyber Security Centre's small business guide covers the basics around it. See how it works.

Only if you say it may, and only up to the limit you set. Before setup you choose, job by job, whether the assistant drafts for you to approve or acts by itself - for example, replying to booking enquiries on its own but never paying a bill. Spending and deleting have their own limits. The rules are written down, and you can tighten them at any time. The steps of a setup show where you set them.

It sees only what you connect it to, so it can do the job you set. An agent that drafts your posts needs your content, not your customer list; one that makes invoices needs your job records, not your inbox. We connect the least access each job needs, inside your own accounts. If customer details are involved, the ICO's guidance on AI and data protection explains your duties in plain terms. Ask us about your case on the contact page.

Yes, UK GDPR allows it, as long as you have a lawful reason, tell people how you use their data and keep it secure - the same rules as any other software. Keeping the assistant inside your own accounts, with written limits, makes those duties easier to meet and to show. The ICO's advice for small organisations has free, plain guidance and checklists. This page is guidance, not legal advice. Ask us if you are unsure.

It depends on the plan and the provider, so it is worth checking before setup. We prefer business plans whose providers say they do not train on your conversations by default, and the setting sits in your own account for you to check. Each provider sets out its own position - Anthropic, for example, explains how it uses personal data in model training. Read more about who sets it up and how we choose tools.

Its limits keep a mistake small: it cannot send, pay or delete beyond what you allowed, so the worst case is one you already agreed to. For anything that matters, you can keep it on drafts for you to approve. Overnight agents run behind a safety lock that stops when it is unsure and leaves you a morning report - see Autonomous Agent Operations.

Yes, at any time and without asking us. Because you add our access through each tool's own team or delegate settings, you can switch it off in the same place. Everything we set up stays in your accounts after that - your data never lived anywhere else. Monthly services can be cancelled any month. If you would like help tidying access at the end of a job, message us. The NCSC's guide to securing your important online accounts is worth a read.

Any online account can be, so an AI account needs the same care as your email: a strong password and two-step verification switched on. The AI is only as safe as the accounts it is connected to. We never ask for passwords by message, and we connect each assistant with the least access its job needs, which limits the harm if one login is ever lost. The NCSC's advice on 2-step verification shows how to turn it on. See what every service includes.

You can, at any point, and nothing is lost. One-off setups are a single payment with nothing to renew, and monthly services can be cancelled any month. Switch off the connections in your tools' own settings, and your emails, files and records stay exactly where they always were - in your accounts. You are never tied to a tool of ours. The prices page shows which services are monthly, and the contact page is there if you want help switching off.

We follow the playbook large firms use, sized for a small business: start with the smallest blast radius, build in the open inside your accounts, and widen access only with sign-off. It matches the UK government's AI Cyber Security Code of Practice and the OWASP guidance on excessive agency - least access, and a person approves anything high-impact. The people behind Eclipse Labs AI learned it inside PwC, Wipro and tech unicorns. More on how it works.