Privacy Policy
What Tium collects, what it never stores, and exactly which third parties can see a request. If you only read one section, read the second.
Effective 8 October 2026
Who this covers
This policy covers tium.ai, the Tium dashboard, and the inference API served from our gateway. Tium is the controller for your account data and the processor for the requests you send through the API.
It does not cover whatever you build with the API. If your own application collects personal data from your users and passes it to us in a prompt, you are the controller of that data and this policy describes what we do with it on your behalf.
We do not store your prompts
The gateway reads your request, forwards it to the model that answers it, and streams the response back to you. It does not write the request body or the response body to a database, a log, or a file. When the connection closes, the only thing left is a row of numbers.
That row is:
- a request id, for support correlation
- the model requested
- input, cached, and output token counts
- the weighted-token total and the multiplier applied
- time to first token, total latency, and a status code
- the API key the request authenticated with
- how long the request waited for a free slot (slot keys only)
- how the response ended: finished, hit the length limit, called a tool, or was cancelled
- the reasoning token count, where the model reports one
- the number of tool calls in the response, never their content
This is a property of how the gateway is built, not a promise about how carefully we delete things: there is no deletion step because there is no write. It is also the reason we cannot help you recover a response you lost: we never had it.
The important limit: while your request is being answered, its full content necessarily passes to the inference provider serving that model. What they retain is governed by their terms, not ours. Section four names every one of them.
What we do collect
Account
GitHub account id, username, email address, and the date your GitHub account was created.
Why: To identify you at sign-in, contact you about the service, and, using the account age, to judge whether a new account qualifies for trial credit without being an abuse vector.
Kept: For as long as the account exists, then 30 days after deletion.
API keys
A SHA-256 hash of each key, its visible prefix, and its last four characters. Never the key itself.
Why: To authenticate your requests and to let you recognise your own keys in the dashboard. Because only the hash is stored, we cannot show you a key again after you create it, and neither can anyone who obtains our database.
Kept: Until you revoke the key, then 90 days.
Usage metadata
Per request: an id, the model, token counts, weighted tokens, latency, and a status code. Not the request or response content.
Why: To bill you accurately, to show you your own usage, and to detect metering faults on our side.
Kept: 24 months, which covers the period in which a billing dispute can reasonably be raised.
Billing
Payment processor customer and subscription identifiers, your plan, period dates, an append-only ledger of credits bought and spent, and the country of your card and billing address as the payment processor reports them.
Why: To run subscriptions and credits, and to be able to reconstruct any balance we have ever shown you.
Kept: Seven years, for tax and accounting obligations. This outlives account deletion by necessity.
Site analytics
A random identifier stored in a first-party cookie on your browser, the pages you view on this site, the site that referred you, and any campaign tag in the link you followed. If you sign in, these are linked to your account from that point on, including the visits you made before signing up.
Why: To count visitors and to know which channels bring people who go on to use the service. It stays in our own database and is never sold, shared, or sent to an advertising network. If your browser sends Global Privacy Control or Do Not Track, we withhold the third-party hosting telemetry described above; this first-party record continues, because it is not shared with anyone.
Kept: 24 months.
Approximate location and locale
The country and state or region your connection appears to come from, read from our hosting provider's standard geolocation at sign-up and on later visits, plus your browser's time zone and preferred language, and the page you were signing in to reach. Never your IP address, and nothing more precise than a region.
Why: To plan where we run capacity, to meet tax obligations that depend on where customers are, and to contact you at sensible hours. It is approximate, and we do not use it to make decisions about your account.
Kept: For as long as the account exists, then 30 days after deletion.
Waitlist
Email address, an indication of how much you use agent tooling, and the referring page.
Why: To tell you when access opens and to understand which channels bring people who actually sign up.
Kept: Until you ask to be removed, or 24 months after launch.
We do not buy data about you, we do not sell or share your data for advertising, and we run no advertising trackers. Traffic measurement on the marketing pages is cookieless and aggregate: it carries no account identifier and cannot be joined to your usage.
Who else can see your data
Slots: your requests are processed only on servers we rent and operate in the European Union. They are never sent to a model lab or any other inference provider.
Pay-as-you-go API: your requests are answered by the lab that built the model you chose, which may process them outside the EU, including in China.
This is the complete list of everyone who can see your data.
Slots
| Party | Role | Region |
|---|---|---|
| GPU servers we rent and operate | Runs GLM-5.3 for slot customers. We operate the software; the hosting company provides the hardware.The content of slot requests and responses, in memory only, for the duration of each request. Nothing is written to disk or logged. | European Union |
| Our gateway server | Checks your key and meters usage before a request reaches the GPU serversYour request in transit, which it passes on without storing. It keeps your key hash and usage counts; no request content. | European Union (Germany) |
Account, billing, email and sign-in services are shared with the API and listed below. None of them sees request content.
Pay-as-you-go API
| Party | Role | Region |
|---|---|---|
| Model-lab inference APIs | Runs the model that answers your pay-as-you-go API request, at the lab that built itThe full content of your request (messages, system prompt, and any tool definitions) for the duration of the request, plus the response streamed back. | Outside the EU, including China |
If routing your requests outside the EU is unacceptable to you, use slots, or write to privacy@tium.ai before you integrate and we will tell you which endpoint currently serves which model.
Shared by both
| Party | Role | Region |
|---|---|---|
| Database hosting | Managed Postgres: the account, billing, and usage databaseYour account record, API key hashes, usage metadata, credit ledger, and slot reservations. No request content. | European Union (Germany) |
| Application hosting | Application hosting, plus cookieless traffic and Web Vitals telemetryHTTP request metadata for pages served, and aggregate performance measurements that carry no account identifier. No request content. | United States / global edge |
| Payment processor | Payment processing and invoicingYour payment details, which you enter with the payment processor directly, plus billing email and the amounts charged. For slots, also your company name and the invoiced reservation. We never receive or store card numbers. No request content. | United States |
| Sign-in provider | Sign-in (OAuth)An authentication request. Returns your account id, username, email, and account creation date to us. No request content. | United States |
| Email delivery | Transactional email deliveryYour email address and the contents of service emails: account and balance notices, and slot reservation, confirmation and invoice emails. No marketing email is sent through it. No request content. | United States |
| Business email | Our mailboxes, for mail you send us and our repliesWhatever you write to us, and our replies. No request content. | United States |
| Bot-check service | Tells people from bots on our public formsYour IP address and browser signals when you submit a public form, such as a slot reservation. No request content. | Global |
We will publish a change to this list before a new provider begins receiving traffic, not after. If a change is unacceptable to you, you may cancel and request a refund of unused credits under section three of the Terms.
International transfers
Tium is established in the United States, but your account data (your profile, keys, usage records, and credit ledger) is stored in the European Union. If you are in the EEA or the United Kingdom, that data does not leave it.
Our gateway (the server your requests actually connect to) also runs in the European Union. It holds your API key hashes and a short-lived queue of usage records; it does not hold request content.
For slots, request content stays in the European Union, on servers we operate. For the pay-as-you-go API, request content goes wherever the model you chose is served from. That may be outside the EEA, including China. It is not stored at either end by us, but it is transmitted, and section four says exactly to whom.
Jurisdictions differ in what protection they offer and in what access local authorities may have to data held there. We rely on standard contractual clauses with providers who offer them, and on your explicit choice of model where they do not. Because model choice is yours on every request, you retain direct control over where your request content goes; the model list documents which provider serves which model.
If we emailed you about Tium
We email founders and technical leads at companies that run AI agents, about Tium's capacity for them. If that was you, here is what we hold and why.
We hold your name, role, work email, company, country, and the public fact about your company that our first email mentioned. We found them through business contact research tools and your company's own public pages, and we record which source each address came from. We use them only to tell you about Tium, because it is relevant to your company's work. That is our legitimate interest, and the only purpose.
We send at most four emails and stop as soon as you reply. Reply "stop" to any of them and we delete your details, keeping only your address on a do-not-email list so you are never contacted again. If you never reply, we delete your details 12 months after our last email. Every email is read and approved by a person before it is sent, and none of them track opens or clicks.
You can ask what we hold, object, or complain to your national supervisory authority. Reply to our email or write to privacy@tium.ai.
Your rights
You can see your account data, correct it, export it, or delete it. Most of it is already in the dashboard: your keys, your usage, your ledger.
Deleting your account removes your profile and revokes your keys immediately. Two things survive, and you should know which: usage rows are kept for 24 months so a billing dispute can be settled, and billing records are kept for seven years because tax law requires it. Neither contains prompt content.
If you are covered by the GDPR or UK GDPR you also have the right to object to processing, to restrict it, and to complain to your national supervisory authority. If you are a California resident you have the rights to know, delete, correct, and opt out of sale or sharing, but we do neither, so the last one has nothing to act on.
Write to privacy@tium.ai from your account address. We answer within 30 days and we do not charge for it.
Security
API keys are stored only as SHA-256 hashes. Someone who obtained our database could not use them, and neither can we: that is why a key is shown once and never again.
Card details never reach our servers; our payment processor collects them directly and we hold only their customer identifier. Database access is restricted to the application's service credentials, and every user-facing table has row-level security enabled with no public policy, so nothing is reachable with a public key.
No system is perfect. If we discover a breach affecting your personal data, we will tell you and the relevant regulator within 72 hours of becoming aware of it, and we will tell you what actually happened rather than the smallest true thing we can say.
Children
The service is not directed at anyone under 16, and we do not knowingly collect their data. If you believe a child has created an account, write to privacy@tium.ai and we will remove it.
Changes to this policy
When something material changes (a new subprocessor, a new category of data, a narrowed right), we update the effective date at the top and email account holders before it takes effect. Cosmetic edits do not move the date, because a date that moves for typos is one nobody reads.
Questions about any of this: support@tium.ai.