Privacy policy
This policy explains how Incremento Ltd ("Spring Prompt", "we", "us") collects, uses and protects information when you use this website and our services. We are the controller of that information. Incremento Ltd is registered in England and Wales, company number 12997528, registered office 128 City Road, London EC1V 2NX.
What we collect, why, and our lawful basis
- Messages you send us. If you use the contact form, we receive your name, email address, company (if you give it), the topic, your message and the page you came from. We use them to reply and to discuss working together. Lawful basis: our legitimate interest in answering enquiries, and taking steps you ask for before entering into a contract.
- The weekly email. If you sign up for the weekly “launches and retirements” email on our model retirements or release tracker pages, we store your email address, which page you signed up on, the words you agreed to, when you signed up, confirmed and unsubscribed, when we last sent you an issue, and two random codes for the confirm and unsubscribe links. We use them only to send you that email. Until you confirm by opening the link we send, the only email you get from us is that link. Every issue has a link to unsubscribe in one click, or you can email hello@springprompt.com. After you unsubscribe we keep your address marked as unsubscribed, so we don't email you again; ask and we'll delete it. We don't add open or click tracking to these emails. Lawful basis: your consent, which you can withdraw at any time.
- The online masterclass. If you sign up for the masterclass, we store your name, email address, company (if you give it), what you say you are building (if you say), which session you could make, your browser's time zone, the link you came from, the words you agreed to and when you signed up. We use them to send you the joining link and the slides, to tell you if the time changes or when the next one is, and to plan the sessions. Lawful basis: taking steps you ask for, and our legitimate interest in running the event. We delete them when you ask: email hello@springprompt.com.
- Client data. If you engage us, the product data, images and outputs you share are handled under our agreement with you. They are stored privately and never used in our public benchmarks. Lawful basis: contract.
- Signing in with Google or GitHub. If you choose "Continue with Google" or "Continue with GitHub", we send you to that company to sign in, and it tells us who you are. Google sends us your email address, whether Google has verified it, and Google's own identifier for your account. GitHub sends us your public profile, which includes your username, your name if you have set one and the date your GitHub account was created, and the email addresses on your GitHub account with which of them are verified. From all of that we keep only your email address (the one Google has verified, or your verified primary address on GitHub) and a note of which of the two you used. We read the age of a GitHub account once, to decide whether a new account can start a free trial straight away, and do not keep it. We do not keep your name, username or profile, and we never store the access token either company gives us: it is used for that one sign-in and dropped. We cannot see your password, your Gmail or your repositories, and we ask for no permission to. While you are being sent there and back, a cookie,
sp_oa, holds a random code for up to ten minutes so that the answer can only be used by the browser that asked. Signing in this way is optional if you already have an account or an invite: you can always use a code sent to your email instead. Lawful basis: taking steps you ask for before entering into a contract, and our legitimate interest in stopping one person starting many free trials. - Your account and the signed-in tools. We hold your email address to sign you in, and a record of any invite you used. If you started a free trial yourself, we also keep when you did and a scrambled form of the IP address you signed up from (a keyed hash, which can't be turned back into the address without a secret only our server has). We use it only to limit how many free trials start from one network. What you enter in Prompt tests, the Playground and any other signed-in tool, including prompts, any shared context (documents or briefs you give a test to send with its cases), test cases, checks, ratings and uploaded feeds, is stored privately, with the test. If you share a rating link, anyone with it can read that test's inputs and answers, and the name and ratings they give are stored with the test. Keys for coding agents are stored only as a hash. It is sent to the AI model providers you choose (through OpenRouter) and to the models we use to judge answers, draft test cases and improve prompts. On the Pro plan, by default, we ask OpenRouter to send it only to providers that keep none of it once they have answered (OpenRouter's zero data retention), so the tools offer only models that have such a provider; a Pro account can switch to all providers in its settings. On other plans, model providers may keep what is sent under their own policies, and such models are marked. Your email address is not sent to them. Tests and results are deleted after 90 days, or sooner if you ask. A test you have switched scheduled rechecks on for is the exception: it is kept, with its results, for as long as they stay on and for 90 days after they go off. Results are private unless you create a share link, which you can switch off. Please don't enter other people's personal data unless you have the right to. Lawful basis: contract, to provide the tools you signed up for.
- Trying the Prompt test before you sign up. If you paste a prompt on the Prompt test page without an account, we send it to an AI model provider (through OpenRouter) to draft test cases and checks for it. We keep the prompt and the draft for 24 hours, so they are waiting if you sign in, and then delete them. A cookie,
sp_pv, lets your browser find the draft for those 24 hours. To limit how many drafts one network can ask for, we keep a scrambled form of your IP address (a keyed hash) with it for the same time. The prompt and the draft are not sent to Google Analytics, are not written to our logs, and are masked in session recordings. Lawful basis: our legitimate interest in showing you the tool before you sign up. - Conversation tests. In a conversation test, a model plays the user: ours by default, or one you choose. Each scenario you write (who the user is, what they want, what they know and any first message) is sent to that model, through OpenRouter, with the conversation so far, so that it can write the user's next message. The agent's prompt, the tools you describe and the results you write for them are sent to the models you choose to test. Those tools are dummies: nothing is called or changed outside the test. Each conversation's transcript, with its tool calls and their results, is stored with the test's results, read by the judging model, and kept and deleted like any other result. Lawful basis: contract, to provide the tools you signed up for.
- Payments and credit. When you buy credit, Stripe takes your payment and holds your card details. We never see your full card number. We keep a record of your purchases, your credit balance and what each run cost. Credit you buy ends 12 months after you add it; about 30 days before some of it ends, we send you one email saying how much ends, on what date, and your balance. If you take a plan, Stripe holds the subscription and takes each month's payment. We keep which plan you have, its dates and whether the last payment went through. Lawful basis: contract, and our legal obligation to keep financial records.
- Changes to your saved tests, and checks before a merge. Each Prompt test keeps the time it last ran and the list prices of its models then, so we can show you when a newer model has come out, a model is being retired or a price has changed since. If we email you about such changes (a model alert, at most once a day, which may also say what Live tried by itself on your prompts), we keep a record of which changes it told you about, and if you turn those emails off with the link in any of them, we keep that choice so that it holds. When a check before a merge is run against a test with one of your account keys, the test keeps, for the newest 30 checks, the verdict, the figures, short excerpts of a few test cases with their answers before and after, and for checks that got worse, the reason. The candidate prompt is sent to the model you choose and the judging model, as in any test. If the check runs from our GitHub Action, the same verdict and excerpts are posted in the pull request, in your repository on GitHub. A check of a test on a Live project's real calls keeps and posts no excerpts, only the verdict and the figures. Lawful basis: contract, and our legitimate interest in telling you when a test you saved may be out of date.
- Account keys, the API and scheduled rechecks. If you set a daily spending limit on one of your account keys, we store that limit with the key. When your code calls our API with a key, we record against your account which action was called and when, and, if the call was refused, the reason. These are usage records like the others below, and they never include what the call sent. The contents of a call are kept only where the action itself saves them: a test you start through the API is stored like any other test, and a call that only reads stores nothing more. Credit that a call spends is marked in your credit record with the key that spent it, which is how a key's limit is kept. If you switch scheduled rechecks on for a test, the test keeps that choice and when you made it, the times of its last 10 rechecks, how the last one ended, the reason if one was due and did not run or if we switched rechecks off for you, how many rechecks have run since the test was last opened, and, when a recheck changed the model the test recommends, the date and the model before and after. A recheck sends the test's prompt and cases to the models you chose and to the judging model, as any run does, and the answers it gets replace the earlier ones. Credit it spends is marked as a scheduled recheck in your credit record. A test with rechecks on is not deleted after 90 days like other tests: we keep it, with its results, for as long as rechecks stay on, and for 90 days after you switch them off or they switch themselves off. The recheck settings and history are deleted with the test. Lawful basis: contract, to provide what you switched on.
- Webhooks. If you set a webhook on your keys page, we store the address you give us and a signing secret we make for it. The secret is stored encrypted, with a key only our servers hold, rather than as a hash, because every message has to be signed with it; we show it to you once and never again. When a run of one of your tests ends, or a run changes the model a test recommends, we send that address a small signed message: the test's id and name, how the run ended, what started it, and the test's recommendation as our API gives it (the model, how many cases passed, the cost figures and a short reason). It never includes your prompts, test cases or answers. For each message we keep a record for 30 days: the kind of event, the test it was about, when it was made, how many attempts it took, and the HTTP status your server answered or a short error of our own. The message itself is kept only until it is delivered or we give up on it, which is normally within nine hours, and nothing your server answers is stored. Removing the webhook deletes the address, the secret and those records. The address is yours to choose, and what happens to a message once it arrives there is up to whoever runs it. Lawful basis: contract, to provide what you switched on.
- Live: your app's real calls. If you add our package to your app, or post your app's model calls to us yourself, we receive a sample of the calls it makes to AI model providers: the address and path it called, what it sent and what came back, how long it took, the file and function in your code that made the call, the name you gave the prompt if you did, which release and environment it came from, and a count of all calls whether sampled or not. We never receive the headers of those calls, so never your provider keys, nor anything after a
?in the address, nor images, audio or files. Anything that looks like a key or token is removed in your app before it is sent, and again by us before anything is kept; email addresses, card numbers and IP addresses are removed too before anything is kept, unless you switch that off in the project's settings, where you can also have phone numbers removed; and you can have any of the four removed in your app, before they are sent. What arrives is stored encrypted, with a key only our servers hold, and deleted after the project's retention: 30 days unless you choose 7 or 90, counted from when each call arrived (a shorter retention applies at once to calls already kept). When a prompt is tested, one model, through OpenRouter, is shown a sample of its calls with keys, email addresses, phone and card numbers and IP addresses removed, and writes made-up cases like them; only those made-up cases are sent to the models being compared and to the models that judge the answers, as in any Prompt test. If you choose to test on real calls instead, those calls are sent to the models being compared and the judges. Nothing from them is kept longer than the calls: when the earliest call a test used reaches the project's retention, the test's cases, answers and anything written about them are deleted, and only its scores, costs and times stay. Such a test can't be shared or rechecked on a schedule. A prompt is named by your app, by the function that calls it, or by its file, never by words from its calls. We keep counts by day (calls, errors, time taken and tokens) for as long as the project exists, and a timeline of what changed and what we recommended. A prompt's first look also sends its instructions and the checks drafted from them (never a call or a made-up case) to one model, through OpenRouter, which suggests ideas for its wording; the ideas are kept with the prompt. If you set up email or a Slack or Discord address for a project, we send it a short report each morning (or on Mondays, or only when we found something, as you choose in the project's settings): the day's call count and estimated spend, your prompts' names, what we found and what was tried. Each email has a link to stop them. If you connect a computer withspringprompt init, we keep a record of that sign-in for at most 10 minutes: fingerprints (hashes) of its two one-time codes, the short code shown on screen, the computer's name and the project you chose. The project key it gets is made when your terminal collects it, once, and is not kept in that record. Deleting a project deletes everything in it at once, including every test made from its calls. Our database provider's backups hold deleted data for a short while longer before they expire. The calls your app sends may hold your own users' personal data: you decide what is sent, and for that data we act on your instructions as your processor. Please only send what you have the right to. Our Data Processing Agreement sets out our terms as your processor. Lawful basis: contract, to provide what you set up. - Getting-started and product emails. When you open an account we send you a short email on most days of your first week, and a few after it, each with the next thing worth doing; which one depends on what you have done so far in the app (for example whether you have run a test or connected Live), and may include figures from your own account, such as calls seen or a saving found, never your prompts or their answers. After your first week (or sooner, if they are turned on in your email settings) we send "What's new at Spring Prompt" each morning there is something new (or weekly, if you choose): the models that came out, what we shipped, benchmarked and wrote. To learn which emails and versions work best, each account is put at random into versions of these emails (for example with or without a picture), and we keep a record of which version each email was and when it was sent, against your account; we do not add open or click tracking to them. They are sent through Postmark. You can change or stop them at any time on your email settings or with the link in each email, and we keep that choice for as long as we keep your account, so it holds. Lawful basis: our legitimate interest in helping you use what you signed up for, and, for product news, the rules on sending marketing to existing customers, which let you stop it at any time.
- Spending limits and the weekly spend alert. If you set your own spending limits or a weekly spend alert on your credit page, we store them with your account. If you set an alert, we check once an hour how much credit you have used since Monday, from the record of what each run cost, and the first time in a week it reaches your amount we send you one email saying so: the total, what it went on (tests in the app, your coding agent, the API, scheduled rechecks and checks before a merge), and the credit you have left. We keep the week we last emailed you and when, so you get at most one a week. Clearing the alert stops the emails. Lawful basis: contract, to provide what you switched on.
- How you use the signed-in tools. To see where the tools work and where people get stuck, we record what you do in them against your account, such as pages and steps reached, runs started and finished, their cost and time, scores, and errors you see. These records never include your prompts, test cases or answers. They are deleted after 12 months. Lawful basis: our legitimate interest in improving the tools.
- Feedback. If you send feedback, or answer a question in the tools, we keep what you write with your email address, the page you were on, the last error and your browser, and email it to ourselves so we can reply. We only look at a test's contents if you tick the box allowing it. Lawful basis: our legitimate interest in improving the tools and answering you.
- Website analytics. We use Google Analytics to count visits and see which pages are read, where visitors come from and what device and browser they use. We also count contact-link clicks, successful contact submissions by topic, and use of public interactive demos. Google receives your IP address, a cookie identifier and details of the pages you view; it does not receive what you type into the site. To keep automated programs out of the figures, Google Analytics is loaded only once you have moved the pointer, touched the page or pressed a key, or have had the page in view for 20 seconds; a page that is opened and closed untouched sends Google nothing. Lawful basis: our legitimate interest in understanding how the site is used so we can improve it.
- Which pages lead to sign-ups. To see which pages and buttons lead people to start a free trial, ask for an invite or a hand setting up, or run a first test, we keep two cookies on our own site. The first records the first page you opened on a visit, the name of the website that linked to it (its address only, such as google.com) and any campaign tags in the link (
utm_source,utm_medium,utm_campaign), with the day. The second records which of our buttons you last pressed and the page it was on. When you press one of those buttons, or sign up, sign in, send a request or start a test, we store those details with that event, against your account if you have one. They contain no name, email address or identifier of their own. We also send the button and page to Google Analytics with sign-up and request events. If you have an account, Google Analytics is also told, once each, that the account was created, that its first test finished (with the number of models and examples, never their content) and that test credit was bought or a plan month was paid (the amount and a scrambled reference for the payment, never your card, name or email address). If a link with campaign tags first brought you to a page that does not load analytics, such as an invite link, the tags are passed to Google Analytics on the next page that does. Lawful basis: our legitimate interest in knowing which pages are useful enough to lead people to try the tools. - Session recordings. To see where the site is confusing, especially signing up and running a first test, we record how visitors use its pages: which pages you open, where you click and scroll, and how the page changed on screen. We build and run this ourselves; the recordings are stored with Supabase and no analytics company receives them. Anything you type is masked before it leaves your browser. In the signed-in tools and on the sign-in pages the page text is masked too, apart from short labels on our own buttons and headings, so your prompts, examples, answers and results are not recorded. Shared result links are never recorded. If you are signed in, a recording is linked to your account. Only Spring Prompt can watch them. Recording starts at the same moment as Google Analytics does: once you have moved the pointer, touched the page or pressed a key, or had the page in view for 20 seconds. If your browser sends a Global Privacy Control or Do Not Track signal, we do not record you. Lawful basis: our legitimate interest in finding and fixing what gets in people's way.
- Server logs. Our hosting provider records technical request data, such as IP address and pages requested. Lawful basis: our legitimate interest in keeping the site running and secure.
We also process information where the law requires us to, for example to keep business records.
Spring Prompt Live: what people ask
Where is my data stored?
In the United Kingdom (London), with Supabase, encrypted with a key only our servers hold. Our web server, which every call passes through on its way there, runs in the EU (Amsterdam, the Netherlands), with Railway.
Who else sees it?
The AI model providers a test uses, reached through OpenRouter. On the Pro plan, by default, only ones that keep nothing once they have answered (zero data retention). On other plans, model providers may keep what is sent under their own policies, and such models are marked. They see made-up cases, not your calls, unless you choose to test on real ones. Everyone we use is listed under Service providers below.
Do you train on it or sell it?
No. We don't sell your data or use it to train AI models.
Can I remove more before it leaves my app?
Yes. Pass a mask list to remove email addresses, phone numbers, card numbers and IP addresses in your app, or a redact function of your own. Both run before anything is sent. What is sent, and what never is
How do I stop it?
Switch collecting off in the project's settings, or remove SPRINGPROMPT_KEY: without a key the package does nothing at all.
How long is it kept?
For the project's retention: 7, 30 or 90 days from when each call arrived, as you choose. Tests on made-up cases are kept for 90 days like any test. A test on real calls loses its cases and answers when its calls go, and keeps only its scores, costs and times.
How do I delete everything?
Delete the project. Everything in it goes at once, including every test made from its calls.
Do you have a DPA or SOC 2?
For your users' data we act as your processor, under our Data Processing Agreement, which applies to every account; ask us for a signed copy. We have no SOC 2 report yet, and would rather say so than imply one.
Cookies and storage on your device
Browsing this website sets Google Analytics cookies (_ga and _ga_*), which give your browser a random identifier so repeat visits can be counted; they last up to two years. You can block or delete them in your browser settings, or install Google's opt-out add-on. Our own pages set sp_ft, the first page of your visit, the site that linked to it and any campaign tags, for 30 days, and sp_lt, the last of our buttons you pressed and its page, for a day (see which pages lead to sign-ups), and sp_next, the test you were starting when you signed in, for a day, so you land back on it. If you sign in with Google or GitHub, sp_oa holds a random code for up to ten minutes while you are away (see signing in with Google or GitHub). If you hide the free trial banner, your browser's local storage keeps sp_offer_off, the time to show it again, for a week. If you have an account, local storage also keeps a short note (names starting sp.ga.) once Google Analytics has been told that your account was created, ran its first test or bought credit, so it is not told twice. If you sign in, we set one strictly necessary cookie, sp_session, which keeps you signed in; it holds no tracking data and is removed when you sign out or it expires. Session recordings keep a random identifier in your browser's local storage (sp_rv) so visits from the same browser can be put together, and the current visit in session storage (sp_rs), which ends when you close the tab. Session storage also keeps sp_h, a note that someone has used this tab (it holds only the figure 1), so that analytics and recording start at once on the next page you open; it too ends when you close the tab. A cookie of the same name, also holding only 1, tells our server that a button you press was pressed by a person rather than a program, so it counts the click; it ends when you close your browser. The signed-in tools also keep unsent drafts in your browser's local storage, so a reload doesn't lose your work; that stays on your device. We do not use advertising or cross-site tracking.
How we use it
- To reply to your message and discuss working together;
- To send the weekly email, if you asked for it;
- To deliver the services you engage us for;
- To keep the site secure and prevent abuse;
- To meet our legal obligations.
We do not sell personal information, and we do not use your data to train AI models.
Service providers
We use a small number of processors who act on our instructions:
- Supabase stores contact messages, weekly email sign-ups, accounts, uploads, results, Live calls and session recordings (in the United Kingdom). Contact form messages are stored there for us to read; they are not passed to an email service.
- Postmark sends sign-in and invite emails, the weekly email, product emails and Live reports (which name your prompts, never quote your calls), and passes feedback to us.
- Stripe takes payments.
- Google Workspace holds email you send us.
- Railway hosts the website and our web server, in the EU (Amsterdam, the Netherlands). Everything you send us, including Live calls, passes through it in memory on its way to Supabase.
- Google provides Google Analytics, and may process analytics data in the United States.
- Google and GitHub confirm who you are if you choose to sign in with either. They act for themselves there, under their own privacy policies, and learn that you signed in to Spring Prompt.
- OpenRouter and the AI model providers it routes to process what you run through the signed-in tools, and the data we run in an engagement.
International transfers
OpenRouter, Google, Stripe, Postmark and many AI model providers are based in the United States, so what you send through the signed-in tools, and client data we run through models in an engagement, may be processed there. We send them only what each test needs, never your contact details, and, on the Pro plan, AI model providers through OpenRouter's zero data retention by default, so they keep none of it once they have answered. On other plans, model providers may keep what is sent under their own policies, and such models are marked. If you would rather a model provider outside the UK did not see your data, tell us before we start and we will agree which models to use.
How long we keep it
We keep contact messages only as long as needed to reply and follow up, tests, uploads, results and judges trained on them for 90 days (a test on a Live project's real calls loses its cases and answers sooner, at the project's retention; a test with scheduled rechecks switched on, and its results, for as long as they stay on and for 90 days after they go off), usage records for 12 months, session recordings for 30 days, weekly email sign-ups while you're subscribed, and afterwards only as a record that you unsubscribed, feedback until we have dealt with it, payment records for as long as the law requires, and client data for as long as our agreement requires. You can ask us to delete your information at any time.
Your rights
Under UK data protection law you can ask to access, correct, export or delete your personal information, and object to or restrict how we use it. Email hello@springprompt.com. You can also complain to the Information Commissioner's Office (ico.org.uk).
Changes
We may update this policy. Changes are posted here with a new date.