PRIVACY
What we do with your data.
The question you send passes through us and is not written down. The conversation you have with Clint is kept, scrubbed and short-lived. We keep the bookkeeping and your account. We never train anything on your words. Some of the services that answer do keep them, and we publish exactly who.
How to read this
Drafted by the engineering team so that every factual statement traces to running code. Not reviewed by a qualified data-protection lawyer. Last checked September 13, 2026. Where we cannot yet make a commitment, the document says so, dates it, and names what would close it. We would rather you found the gap here than found it yourself.
What would change that: A review is commissioned when the first controller who is not us signs the processing agreement, or at the first paid business customer — whichever comes first. It is deferred rather than skipped: today there is no customer to make these statements to.
What changed in this version: Section 2 names two stores it had left out: the keys for your own AI accounts with the one-time confirmation that they are yours, and the count we keep when you try Clint without an account. It drops the account role, because an account is one person's. Section 4 says your work reaches a company only through your own account with it. Section 6 says how we test models: our own questions, only where the company allows it, results kept internal. Section 9 says money records are kept seven years after closing, then deleted. It also publishes the registered office for legal notices, adds section 11 on California privacy rights (CCPA/CPRA) and Global Privacy Control, and says in section 13 that every surface discloses Clint is an AI. Counsel review remains open.
A printed copy carries this version and its date, and leaves the navigation behind.
1. Who we are
Host.Rodeo is an AI gateway operated by Scaled Minds. Registered office for legal notices: 4030 Wake Forest Rd Ste 349, Raleigh, NC 27609. You can reach a real person at [email protected]. We are the controller for your account, and a processor for anything you send through the service on behalf of your own users.
2. What we hold, one store at a time
A single sentence about "your data" cannot be true of ten different things, so here is each one.
- The request you send for inference. Passed to whichever service answers it and gone from us when the answer comes back. It is not written to our disks on the way through.
- Your conversation with Clint. Kept, so that moving between pages does not restart the thread. Each turn is scrubbed of anything shaped like an email address, a key or a card number before it is written, both the turn and the thread are bounded in size, and the whole thread ages out on its own. A conversation you have not returned to is not a conversation.
- Anything you tell Clint while setting up — where you are, what you are trying to do. Kept short, scrubbed the same way, and used to answer you rather than to profile you.
- Files and artifacts you upload or that Clint produces for you, for as long as you keep them. Deleting one deletes it.
- The bookkeeping. For each request: which services were considered, which one answered, whether it worked, how long it took, what it cost. Tied to an account number, not to your email address.
- Your account: email address, a scrambled password that cannot be turned back, and your sign-in sessions. An account is one person's, so nobody else's details sit inside it. Your Host.Rodeo keys are kept only as one-way scrambles, never the key itself.
- Your own AI accounts, if you connect any: the key or sign-in for each, sealed in a vault and used only to send your own work there, and the one-time confirmation you gave that those accounts are yours, with when you gave it.
- Your machines, if you connect any: what hardware they have, what they can run, whether they are healthy. Never what is on their disks.
- Feedback you send us, scrubbed the same way.
- Trying Clint at the front door without an account. That conversation is kept as a visitor's thread, on the clock in section 3. To share out the free tries fairly we count how many questions came from your connection each day, under a one-way scramble of your internet address rather than the address itself, and drop the count after two days.
A scrub reduces what is exposed. It does not make what is left non-personal, and we do not describe it as anonymous. An account number is a pseudonym, not anonymity, and text you wrote can identify someone even when no name is attached to it.
3. How long each one lasts
| What | How long |
|---|---|
| What you ask, and the answer that comes back A negative claim, so there is no constant to bind it to. It is held by there being no writer: the request body is never handed to a store on the way through — except a conversation your app keeps with us so its next question can follow on, which has its own row in this schedule. | never written down — it passes through and is gone once the answer has reached you. The one exception is a conversation your app asks us to keep so its next question can follow on, which has its own clock in the schedule |
| A conversation your app keeps with us, so its next question can follow on from the last Only apps that talk to us the way OpenAI's newer agent tools do can keep one, and they do unless they say not to; coding tools such as Codex always say not to. Only the same key that saved it can read it back or carry it on, your app can delete it at any time, and closing your account removes it. A message about a crisis is never kept, and neither is the reply to it. | 30 days after it was saved |
| Your conversation with Clint, signed in Scrubbed before it is written, bounded per turn and per thread, and dropped on the next read or write after it expires — an idle thread ages out with nothing having to sweep it. | 7 days after you last add to it |
| A visitor's thread, before there is an account Filed under a one-way digest of a handle that lives only in your own browser. We hold no name for it, and signing up claims it into your new account exactly once. | 24 hours |
| What you told Clint while setting something up | 7 days |
| The notes Clint keeps so a follow-up makes sense Tell Clint something while you are setting up — keep this one private, bill it to the ranch — and he writes a short note so the next thing you say still makes sense. Every time he reads or writes one, anything older than this goes; a note is dropped for you even if you never come back. | 30 days |
| Activity → Changes: what changed on your account, who changed it, when, and how to undo it One line for each change Clint or you made: what changed and to what, who did it and when, so you can look back over a year and undo what can be undone. At most the latest 500 changes are kept, older ones go on every write, and the whole list is erased when you close your account. | 400 days |
| Files and artifacts you uploaded or Clint produced Backup copies expire on their own 7-day clock, and the record that an artifact existed at all is kept 15 days so a deletion can be proved rather than assumed. | 7 days |
| The record of a task Clint ran for you | 7 days |
| Your sign-in | 7 days, then you sign in again |
| Notices waiting for you in your own ranch | 30 days |
| The detailed record of how one piece of work was handled Which places were tried for you, which one answered, what each attempt took and what it cost. Never what you wrote and never what came back — neither is written here at all, and a test asserts it. Filed under an account number rather than your email address. It is how we can tell you what actually happened when something goes wrong. | 14 days |
| Feedback your app sent us about an answer Whether an answer was good or bad, the reason from a short fixed list (for example 'the answer took too long'), at most a few words of your own, and which model and request it was about. Never what you wrote and never what came back, and a note that looks like either is thrown away. Filed under an account number rather than your email address, with a one-way label for the network you sent it from so no one sender can pose as many. Closing the account removes it. | 30 days |
| What we promised about each answer, and how it turned out How long we said an answer would take and what it would cost, beside what actually happened, so we can publish how often our promises held. Filed under an account number rather than your email address, never anything you wrote, and removed when you close the account. | 400 days |
| What we remember about a searchable collection your app built, so later searches keep matching it When your app makes documents searchable through us, we note which model did the work and the settings that shape the result, so every later search is answered the same way and never quietly stops matching. That is all we keep: never your documents and never what came back. A setting longer than a short word is kept only as a one-way fingerprint. Closing the account removes it. | 2 years after it was last used |
| The bookkeeping behind your Trail and your receipts Which services were considered, which answered, whether it worked, how long it took and what it cost — filed under an account number rather than your email address. | 365 days |
| Your account, your keys, and the machines you connected Closing the account revokes every key immediately and removes the account record. | Until you remove them or close the account |
| What you paid, were refunded and were invoiced, after you close your account Charges, refunds, invoices and what a supplier billed for your work are accounting records. When you close your account they are filed under a random reference instead of your email address, kept seven years from the end of the month they belong to or from the day you closed, whichever is later, and then deleted automatically. A charge still settling when you leave is kept until it settles first, and one that cannot settle on its own is kept until a person resolves it; the seven years run from then. A supplier's statement of what your work cost is counted from the end of the period it covers, or from when it reached us if that is later. After the seven years we keep only the payment processor's customer number, so a late message about an old payment can never land on a new account at your address. | 7 years for the accounts, then deleted |
| A video you asked Clint to make We keep which video you asked for, from which supplier and when, so you can come back and fetch it. The clip itself is held by the supplier, not by us. If a paid video's charge has not settled yet, that one record is kept until it does — so nobody is billed twice and nothing is written off silently. | 30 days |
"Ages out" here means it stops being returned AND is dropped from the store — both are enforced when the row is next read or written, so an idle thread expires without anything having to sweep it. Backups are a separate clock and we do not claim an instant physical erasure everywhere; if you need the exact position for your own record, ask and we will tell you what is true on the day.
4. Who else sees it
To answer your question we have to send it to something that can answer it. That is the whole service and it is the part worth understanding. When you are signed in, that is always one of your own accounts with an AI company, or a machine of yours, and the company holds what you sent under the terms you accepted with it directly. Questions asked at the front door without an account go to Host.Rodeo's own accounts, only with companies whose terms allow a product like ours to use them. The complete current list of every company involved is generated from the running system rather than from a document somebody remembers to update.
For each one we say where it is based, whether it keeps what you sent, and whether it learns from it. Where a company has not published a clear answer we say not known, rather than guessing in our own favour.
Two things from that page deserve saying here as well. Some of the companies whose free accounts you can connect do learn from what you send them, and say so in their own terms. And some of the services are themselves middlemen who choose another company per request without telling us which — for those we genuinely cannot name the final destination in advance, and saying so is the only true answer available.
5. How to make sure nobody else sees it
Mark the work private. Private work only ever runs on machines you own. If none of yours can take it we refuse the request and tell you — we never quietly send it somewhere else. Every answer comes back with a signed receipt naming who handled it, and you can check that signature yourself without trusting us. It is the one guarantee in this document that is a mechanism rather than a promise.
6. Training
We never train, tune or test any model on your data. Nor do we let your words judge how good a model is — that comes only from our own frozen test set, so one person's activity cannot shape another person's answers. We do test models, with our own questions only, and only where the company offering the model allows it in its terms; the results decide where requests go, stay with us, and are never published as a comparison that names anyone. We do notice when a service breaks and route around it, but that is read from the service's behaviour, never from what you wrote.
What we cannot promise for you is what a third-party service does once your words reach it. That is why section 4 exists, and why private routing exists.
7. Where your data goes
Our own servers are in the United States. The services that answer questions are mostly in the United States, with some elsewhere, and a few whose location we cannot establish. Most do not publish where the computer running the model physically sits, so we report that as not known instead of inventing an answer. The live list shows the current position for each one.
If you are in the UK or the EU, using this service means your data leaves the UK and the EEA. For businesses, the transfer terms are in the processing agreement. If you need a guaranteed location, use private routing on a machine in the location you want — it is the only mechanism that actually guarantees it.
8. Why we are allowed to do this
- To give you the service you asked for: running your requests, keeping your account, keeping your conversation, showing your receipts.
- Because we have to: keeping enough of a record to bill honestly and to answer a regulator.
- Because it is in our legitimate interest, balanced against yours: keeping the service secure, stopping abuse, and knowing when something is broken.
We do not run advertising, we do not profile you, we do not sell or share your data, and nothing here makes an automated decision with a legal effect on you.
9. Your rights
You can ask us for a copy of what we hold, to correct it, to delete it, to limit what we do with it, or to object. If you are signed in you do not have to ask at all: your account page hands you the whole file, now, with a list of every place we looked. There is no charge and you do not need a working account — these rights outlive the account, which is why the contact page needs no sign-in.
Closing your account removes the account record, revokes every key, and deletes your settings and usage totals. The bookkeeping ages out on the schedule above; ask and we will remove it sooner. We will not delete a record we are legally required to keep — an unresolved payment obligation, for instance, or the money records the accounts need, which we keep for seven years after you close the account and then delete — and if that applies to you the deletion receipt says which record, why, and when it goes.
If you are unhappy with how we have handled your data you can complain to your data-protection regulator — in the UK, the Information Commissioner's Office.
10. Security
Passwords are stored scrambled with a slow, deliberately expensive algorithm. Keys are stored as one-way scrambles. Sign-ins expire. Secrets live in a sealed vault rather than in the code. Everything travels over encrypted connections, and the administrative parts of the service are not reachable from the public internet at all.
The full list, including what we have not done, is in Annex II of the processing agreement. We would rather you saw the gaps from us than found them yourself.
11. California privacy (CCPA/CPRA) and Global Privacy Control
If the California Consumer Privacy Act (as amended by the CPRA) applies to you, you have the right to know what personal information we collect, to delete it, to correct it, and to opt out of sale or sharing. Host.Rodeo does not sell personal information and does not share it for cross-context behavioural advertising. We do not use sensitive personal information to infer characteristics about you beyond what is needed to run the service you asked for.
We honour browser Global Privacy Control (GPC) signals as a valid opt-out of sale/sharing. Because we do not sell or share for advertising, a GPC signal does not change our processing today — it is still recorded as your preference and will constrain any future advertising or analytics we might add. You can also exercise California rights without an account via the contact page, or export/delete while signed in at /account/data.
We will not discriminate against you for exercising these rights. Categories we may hold: account contact (email), authentication and session data, API key digests, billing identifiers via Stripe when billing is enabled, machine heartbeats you connect, and limited assistant setup notes. Full inventory: sections 2–3 and /legal/cookies.
12. Children
Host.Rodeo is not for under-16s and we do not knowingly hold data about them.
13. Clint is an AI
Clint is a machine, not a person, and every surface he speaks on says so (including Utah/FTC-style AI disclosure). He can be wrong. He acts only inside boundaries you set, he never commits money without your explicit yes, and anything consequential is shown to you exactly as it will be done before it is done. When he produces an image, audio or video, whatever mark the tool put in the file to say a machine made it stays there — we do not strip it. Each answer can carry a signed receipt naming which model and which company answered it, and what that company has told us about keeping and training on what it is sent.
14. Changes
A new version is posted here with a new number and date, and the previous version stays listed. We tell account holders by email when a change matters. Changes to the list of companies that can receive your data get 30 days' notice.
Earlier versions
Every version this document has had. The one above is what you are reading now.
- Version 1.3 took effect September 29, 2026 — current
- Version 1.2 took effect September 12, 2026
- Version 1.1 took effect August 21, 2026
We do not publish the superseded text, because a stack of old contracts on a public page is a way to be read out of context. The version you accepted stays retrievable: ask us naming the version and we will send you that exact text.
The other documents
- The deal in plain words
- Terms
- Data processing agreement
- Who can receive your data
- Cookies and storage
- Accessibility
Questions about any of this reach a real person at the contact page, with or without an account — [email protected] for anything about your own data, [email protected] for a security report. We answer within 10 working days.