Terms of Service

These terms are between Fluid LLC, an Iowa limited liability company based in Johnston, Iowa ("Fluid", "we"), and the business named in your account ("you"). Your account names the agents you are buying, what they cost, and when they start. If your account and these terms disagree, your account wins on price, agents and dates; these terms win on everything else.

Fluid builds and runs software agents. An agent connects to systems you already own, reads what arrives in them, and does the job described in its playbook — a written instruction we prepare with you and which you can read and change at any time in your dashboard.

Nothing is installed on your systems. Agents run in Fluid's environment and reach your systems through permissions you grant and can withdraw.

Each agent is separate. It can only reach the specific accounts connected to it, and nothing else.

An agent's judgments are made by a large language model. It reads its playbook and the material in front of it and decides what to do.

This means an agent is capable of being wrong in the way a new employee is capable of being wrong, rather than in the way a spreadsheet formula is. It can misread something, misjudge who should handle it, or write something you would not have written. The controls in section 4 exist because of this, and the record in section 6 exists so you can see it happening.

We do not train any model on your data. We contract with our model provider on terms that forbid training on anything we send them, and we will not use a provider that does not offer those terms.

Every agent has a setting you control, which decides how far it goes before a person is involved:

Watching

Works everything out and touches nothing. You see what it would have done.

Prepares only

Does the work and stops before the last step, every time.

Finishes the routine

Completes what its playbook classes as routine, stops on everything else.

Finishes everything

Completes work without you seeing it first.

Each agent starts at the setting agreed for it when it is set up, and that setting is shown in your dashboard. Moving it to allow more is an action you take, deliberately, in the dashboard, and it is recorded.

Regardless of setting, limits apply that you cannot raise: no more than five recipients on any message, no more than ten completed actions in a single run, and no agent is given the ability to delete anything. These are designed into the software, enforced in code, and recorded on every run, so that if one ever failed the record in section 6 would show it.

An honest description of how this is enforced. These limits live in Fluid's software. Google and Microsoft do not offer a permission that distinguishes "may prepare a message" from "may send one", so the distinction is one we make and enforce in code, log, and are accountable for. It is not a restriction the platform imposes on us. We describe it this way deliberately: an agent does not take the final step at the lower settings, and every action it takes is recorded — but you should understand this as a control we operate rather than a physical impossibility.

You can stop an agent at any time. Pausing takes effect on its next run. Disconnecting an account cuts off its access immediately.

What we access. Only the accounts you connect, and only through permissions you grant. You can withdraw them at any time, from your own account settings or from your Fluid dashboard.

What we store. Records of what each agent did, including the text it prepared; the playbook and its history; and metadata about connected accounts. Credentials for connected accounts are encrypted and never leave our servers. We never receive or store your passwords.

What we do not do. We do not sell your data. We do not use the contents of your systems to build anything for anyone else. Data obtained through Google Workspace APIs is used solely to provide the service you have asked for, in line with Google's API Services User Data Policy, including its Limited Use requirements.

Who else is involved. Providing the service requires a small number of subprocessors: our model provider, our database and authentication provider, our hosting provider, our transactional email provider, and the provider of whatever system you have connected. A current list is available on request. We will tell you before adding one that processes your content, and you may object. If you do, we will either find another way or you may end this agreement immediately, with no penalty.

When you leave. Connected accounts are disconnected immediately. Everything is available for export for 90 days, and after that we delete the content your agents worked on. We keep the record of what each agent did — the runs, the settings and the playbook history described in section 6 — for as long as a claim about it could still be brought, and then we delete that too. We keep it because it is the only account of what actually happened, and it protects both of us. We will also keep anything the law requires us to keep, or anything a dispute that has started makes us keep, for as long as that lasts.

You confirm that you are entitled to give us access to the accounts you connect, and that doing so does not breach an obligation you owe to somebody else.

Everything an agent does is recorded and visible to you: every run, every item, the text it prepared, and every change to its playbook and settings. You can export it. This is not a courtesy — it is how you supervise something that makes judgments on your behalf, and you should use it.

There are three kinds of charge, and no amounts in this document. The amounts are the ones shown in your own account, on the billing screen, at all times. They are set out beside these terms on the screen where you accept them, and a copy of those numbers is kept with your acceptance.

• A one-time setup fee — charged once, when an agent is built for you.
• A monthly fee for each agent — flat, and not metered by volume.
• A monthly fee for each person an agent runs for — so an agent that serves four people costs more than one that serves one.

The setting does not change the price. The four settings in section 4 decide how far an agent is allowed to go, and you control that yourself in the dashboard. Moving it costs nothing and changes nothing about your invoice. An agent costs the same whether it prepares work for you to read or finishes it.

Changes take effect on your next invoice. Nothing is charged or refunded part way through a month. If you add an agent or remove one, or the number of people on an agent changes, the amount you pay changes from your next invoice and not before. An agent removed and put back before that invoice costs you nothing extra.

Fair use. There is a volume ceiling per agent. Exceeding it consistently is a conversation about a different tier, not a surprise invoice.

Canceling. You may cancel any agent, or the whole service, at any time, with no notice period. Service continues to the end of the month you have paid for. Fees already paid are not refunded.

Pausing an agent does not stop its fee. Pausing stops it working. Removing the agent is what stops the charge. The charge stops from your next invoice, and the agent stays available to the end of the month you have already paid for. Removing an agent does not delete its history; that is kept and remains available to you under section 5.

Changing prices. We may change your prices at any time by telling you at least 30 days beforehand. If you do not want the new price, cancel — which you can do immediately, with no notice, as set out above. Anything specific we have agreed with you about your prices is written in your account and shown to you beside these terms.

We will provide the service with reasonable skill and care, and we will tell you promptly if something has gone wrong that affects you.

We do not promise the service is uninterrupted or error-free. Agents depend on systems we do not control — the providers you connect, our model provider, the internet. There is no service level commitment in these terms. If you need one, it belongs in a separate written agreement and should be priced.

We do not promise an agent's output is correct. At "Watching" and "Prepares only", a person of yours sees the work before it goes anywhere. At "Finishes the routine", a person sees everything the playbook does not class as routine, and the routine goes without being seen. At "Finishes everything", nobody sees it first. Where that review applies, it is part of how the service is meant to be used.

EXCEPT AS STATED HERE, AND TO THE FULLEST EXTENT THE LAW ALLOWS, WE GIVE NO OTHER WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED, AND WE SPECIFICALLY DISCLAIM THE IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE, AND NON-INFRINGEMENT.

Neither of us is liable to the other for indirect or consequential loss, or for lost profits, revenue or goodwill.

Our total liability for a claim is limited to what you paid us for the agent the claim is about, over the twelve months before the event giving rise to it: that agent's own monthly fee plus one person's monthly fee. Setup fees are not counted, and neither are the fees for any additional people that agent serves.

Nothing in these terms limits liability for fraud, for death or personal injury caused by negligence, or for anything else the law does not permit us to limit.

You are responsible for supervising your agents. You choose the setting, you approve the work at every setting that requires it, and you are responsible for what your business sends.

Each of us will keep the other's confidential information confidential, and use it only to perform or receive the service. Each of us will protect it with at least the care we take over our own, and never less than reasonable care. Either of us may share it with our own people and contractors who need it for this work and who are held to obligations at least as strict as these. This lasts while the agreement runs and for three years after it ends, and for as long as the law protects it where it is a trade secret.

This does not apply to information that is already public without either of us having broken this promise, was already known without breaching a duty of confidence, was worked out independently without using the other's information, or must be disclosed by law. If either of us is compelled to disclose, we will tell the other first where we are allowed to, so they have the chance to object.

Your data is your confidential information. It is also your property: we claim no ownership of it, and no rights over it beyond what is needed to run the service for you.

Fluid's software, playbooks-as-a-format, and everything we build to deliver the service remain ours. A playbook written for your business is yours to keep and take with you.

We may update these terms. Every version is numbered and dated. If we change them, we will tell you in your dashboard and ask you to accept the new version. We keep a record of which version you accepted, when, and from where.

Either of us may end this agreement at any time, as described in section 7. Either of us may end it immediately if the other commits a material breach and does not fix it within 30 days of being told about it.

On ending: your agents stop, connected accounts are disconnected, and section 5 governs your data. Sections 5, 6, 8, 9, 10 and 13 carry on applying after this agreement ends, along with any fees already owed.

Assignment. Neither of us may assign this agreement without the other's consent, except to a successor of substantially the whole business.

Entire agreement. These terms and the details in your account are the whole agreement and replace anything said beforehand.

If a clause fails, the rest still stands.

Governing law and venue. These terms are governed by the laws of the State of Iowa, without regard to its conflict of laws rules. Any dispute goes to the state or federal courts in Polk County, Iowa, and each of us agrees to those courts.

Notices go to the email addresses on your account.