These terms govern the AI assistant built into the EagleHQ dashboard. They supplement our Terms of Service and our Privacy Policy.
This is version 1. If the assistant is given a power these words do not describe, we publish a new version and ask you to agree again. Until you do, the assistant goes back to asking before every change.
What the assistant is
The assistant answers questions about your own business data and can propose changes to it. It runs on large language models operated by third-party providers, and it is a tool inside your dashboard — not an employee, not an agent with authority of its own, and not a person.
What it can see
The assistant can read the records of your site only, and only the ones you could open yourself. It is pinned to your business at the server, so it cannot be persuaded to read another business's data by anything typed into it. Your own permissions apply to it unchanged: a team member who cannot see payroll does not get to see payroll by asking the assistant.
What it can do, and what has to happen first
Every change begins the same way. The assistant writes up exactly what it would do and shows it to you. It cannot write to your records directly — the change is sent from your browser, as you, through the same part of the app that the button would have used. It meets the same permission checks, the same refusals, and the same record-keeping as if you had done it by hand.
What you control is whether you read a card first:
- Ask me every time — nothing changes until you press Confirm. This is the default and it requires no agreement.
- Decide for me — routine changes are carried out as soon as they are proposed. Anything flagged with a warning, anything that reaches one of your customers, and anything that cannot be undone still comes back to you.
- Just do it — anything up to the risk limit you set, without asking.
The last two are delegation, and that is what these terms are for. There is no setting that means "anything at all": a risk limit always applies, a proposal the assistant itself flagged as blocked is never carried out, and your permissions are checked before the mode is even looked at.
What you agree to when you delegate
By switching the assistant out of "Ask me every time", you confirm that:
- You are authorising the changes in advance. A change the assistant makes under that setting is your change. It is recorded against your account, it takes effect immediately, and it has the same standing with your customers, your staff and your accounts as one you typed yourself.
- You accept that it will sometimes be wrong. See the next section. Delegation means accepting that some of those mistakes will happen without you reading them first.
- You will check the log. Every action the assistant takes without asking is written down, with what it did, when, and why it did not stop to ask. Reviewing it is how you catch the ones that were wrong.
- You can change your mind at any moment, and the change takes effect on the next thing it proposes.
We may cap what any account is allowed to delegate, and we may switch delegation off across the platform if we need to. Your settings are kept when we do.
It makes mistakes
This is not boilerplate. Please read it.
Language models produce confident, fluent, well-formatted output whether or not it is correct. This one will, at some point:
- State something untrue about your business, a job, a customer, or a price, in the same tone it uses for things that are true.
- Misread what you asked for and act on the version it understood.
- Act on the wrong record — the right kind of thing done to the wrong job, the wrong customer, or the wrong crew member, particularly where two are similarly named.
- Write text that is inaccurate, off-brand, or unsuitable — including blog posts, descriptions, replies and notes that read well and say the wrong thing.
- Get arithmetic, dates, distances and travel times wrong. Anything it tells you about where a crew member is, who is closest, or how long is left on a job is an estimate drawn from the last position reported to us, and positions can be stale, absent or inaccurate.
- Be wrong about its own limits, including telling you it has done something it has not.
You are responsible for what is published, sent, priced, scheduled and paid from your account, whether you pressed Confirm or delegated the decision. Check anything that goes to a customer, anything that sets a price, and anything that changes money, before it matters. The assistant is provided "as is"; nothing it produces is a warranty by us that it is accurate, and our liability for it is limited as set out in our Terms of Service.
What it will never do
Regardless of your settings, the assistant is not able to:
- Do anything you could not do yourself. It has no permissions of its own.
- Reach another business's data, or the platform's own systems.
- Delete your account, change your plan, move money out, or alter who has access to what.
- See card numbers, bank details, or passwords. We do not put them where it can read them.
What it remembers
The assistant keeps short notes so it does not have to be told the same thing twice — how you like jobs scheduled, what a piece of your jargon means, which supplier you use. You can read every one of them, and delete any of them, from the assistant's settings. Deleting one removes it.
Do not tell it anything you would not write in a business note. It refuses to store things that look like card numbers, passwords, keys, door or alarm codes, or an individual's pay — but that check is a safety net, not a guarantee, and a note you put there is readable by the people at your business who can use the assistant.
Your data and the model providers
To answer a question, the relevant parts of your data are sent to the model provider that serves it. We use providers under agreements that prohibit training their models on what is sent through us. We do not sell your data, and the assistant is not used to build a profile of you or your customers. Model providers are listed on our Subprocessors page, and everything in our Privacy Policy applies.
Conversations are stored in your account so you can go back to them. Delete a conversation and it is gone from the dashboard.
Cost and limits
The assistant is metered in credits included with your plan. Deeper thinking levels cost more per question, and some levels are only available on some plans. If you run out, the assistant stops answering until your allowance resets or you move up a plan — nothing else about your account is affected. There are also short rate limits to stop a runaway loop spending an allowance in a minute; the assistant tells you when it has hit one.
Not professional advice
Nothing the assistant produces is legal, tax, accounting, employment, insurance, engineering or safety advice, and it does not know the licensing rules that apply to your trade in your jurisdiction. Work that requires a qualified person still requires one.
Suspension
We may restrict or withdraw access to the assistant for an account that uses it to generate unlawful or deceptive content, to attempt to reach data it is not entitled to, or in any way that breaches our Acceptable Use Policy.
Contact
Questions, or something it got wrong that you would like us to look at? Email hello@eaglehq.us.
Questions? Get in touch.