Information Security Policy
Access control is covered in section 7.
This policy explains how Algorythyme Ltd protects the information people trust us with. It covers the products we build, the systems we run and the suppliers we rely on. Version 8 takes effect on 17 October 2026 and replaces version 7. We will review this policy by 10 February 2027, or sooner if a significant change to how we work requires it.
1. The company
Algorythyme Ltd is a private company registered in England and Wales (company number 16230652), with its registered office at Unit A, 82 James Carter Road, Mildenhall, IP28 7DE. It runs SteerSpend, a personal budgeting web app at www.steerspend.com, where people upload their bank or credit card statements and the app shows them what is free to spend. The statement files are read on the person’s own device, in the browser, and are never sent to us. To sort payments into categories, short payment descriptions, with names and account details taken out on the device, are sent through our server to Claude, an AI service from Anthropic, unless the person switches this off. Since 13 October 2026, everything about a person’s money has been encrypted in their browser before it is saved, with a key only they can unlock, so we can’t read it (section 5).
SteerSpend cannot move anyone’s money or make payments from their bank account. It does not lend money or give financial advice. Bank connections are read-only: in the US, people can link a bank through Stripe Financial Connections (section 3), and where bank connections launch elsewhere, they will be read-only too. We never ask for or store anyone’s online banking sign-in details.
If we launch a new product, or start collecting a kind of customer information we do not hold now, the rules in this policy still apply. We will record the new information, and its risks, in this policy before it goes live.
2. Responsibility
Algorythyme Ltd is responsible for information security across the company. We keep this policy up to date, approve any new supplier before it gets access to customer information, and are the main contact if there is a security incident.
We check the company inbox, team@steerspend.com, on working days. Customers use the same address, so security questions come straight to us. We do not have a separate security team, and we would rather say so than overstate our size.
3. Information we hold
We keep only what an account needs to work, and only for as long as our retention policy allows. The full list is in the privacy notice, and a summary follows.
Account and sign-in. We hold the person’s email address, display name, how they sign in, when they agreed to the terms, and their country, currency and date format. Passwords are handled by Supabase, and we never see them. Every account must use a passkey or, where a device cannot use one, an authenticator app code. Supabase holds the passkey’s public key, while the fingerprint or face itself stays on the person’s device.
Payments from statements. Statement files are read on the person’s own device and are never sent to us or stored by us. From them, the app keeps only the payments the number needs. For each purchase from about 12 weeks before the current pay period onwards, it keeps the date, the amount, the description as the bank wrote it, a cleaned-up shop name and a category. Payments the app treats as money moved between the person’s own accounts or into savings are kept too, marked as moved. Older purchases are used once, on the device, and are not kept. All of this is stored encrypted.
How payments were sorted. For each payment, we keep the category or type, whether the AI or the app’s own rules chose it, how sure the AI was and its short reason, and the person’s correction, or the fact that they left it unchanged. The most recent 3,000 of these are kept, encrypted. The AI’s suggestion for each of the most recent 1,500 payees is kept in the encrypted account record, so that the same payee is not sent again, along with what was sent to the AI the last 5 times (up to 90 days), for the person to see. For each person, the number of payees the AI sorted each day is kept for a week, to limit use.
What the person sets. We hold the person’s pay details, bills and income, category corrections and rules, cushion, savings targets, the savings they keep track of, loan planner entries, budget adjustments, monthly reviews, notification and display choices, and receipt details if they use that feature. All of this is encrypted, apart from the settings the server needs (country, currency, date format, how the app looks, and whether the AI is switched on).
A copy shared with support. We hold a readable copy of a person’s account only if they choose to share it, and for the time they choose (at most 7 days). It is deleted when that time ends or when they stop sharing. We record each time someone on our team opens it, and keep that record for a year.
Plan status. Stripe handles all card payments, so we never see or store card numbers. We keep the plan status Stripe sends us: whether the plan is in its free month, active, has a failed payment or has ended, its dates, and Stripe’s customer and subscription reference numbers.
Support emails. Support emails are deleted once the matter is resolved, and we review the mailbox at least every three months.
Bank connections (US). In the US, people can link a bank account through Stripe Financial Connections. We keep only which accounts are linked (Stripe’s reference, the bank’s name, the account’s name, the last four digits and the kind), never bank sign-in details. The account’s payments and balance pass through our server to the person’s device, and our server keeps none of them; the app keeps what it needs, encrypted, like a statement’s payments. The bank function uses its own restricted Stripe key, with no access to payments or customers’ cards. The page where people link their bank, bank.steerspend.com, is the only SteerSpend page that loads another company’s script (Stripe’s). It is a separate address, so that script can’t reach the key to a person’s encrypted data, which is kept only for www.steerspend.com; the session it needs is handed to it by message, never in a web address. Accounts are unlinked at Stripe before an account is deleted.
Deleting an account. When someone deletes their account, their login and everything stored with it is removed from the live database straight away, and Supabase’s backups are overwritten after a limited period. People can get a copy of their information before they delete their account.
4. Identifying risks
Section 8 lists the risks we watch. Every six months, we review the list, record any changes and set the date of the next review. We also update it before a new supplier gets access to data, before launching a feature that collects new information, and after any security incident. A report from a customer, a warning from a supplier or a security fix in software we depend on triggers a review straight away, without waiting for the six-monthly one.
5. Protecting information
Every account must use a passkey at sign-in or, where a device cannot use a passkey, an authenticator app code.
Data is stored with Supabase in London, in the UK. The database uses row-level security, so each person can read only their own rows.
Encryption on the person’s device. Everything about a person’s money is encrypted in their browser before it is saved (AES-256-GCM), using a random key made for their account on their device. Each record is bound to its owner and its place, so if it is moved to another account or record, it fails to open. Names in the database (such as a payee for a rule) are replaced by keyed codes that can’t be turned back into names without the key. The key is kept only in forms we can’t use: locked with the person’s recovery code (24 characters, through PBKDF2 with 200,000 rounds), locked with a secret from their passkey where the passkey can give one (the WebAuthn PRF extension), and in their browser, locked with a key the browser keeps and won’t hand out, until they sign out. The recovery code and the passkey’s secret are never sent to us. Each key has an identifier, and the database refuses a record or a key copy made with any key other than the account’s current one. Once an account is encrypted, the database also refuses anything unencrypted for it. What stays readable is listed in the privacy notice. We can’t reset a lost recovery code, but a person with no code, device or passkey that can open their data can start again, keeping their account and plan.
Support access by consent. Our support team can’t read anyone’s payments, but a person can choose to share a readable copy for up to 7 days. Only an admin, signed in with two-step sign-in, can open it. They do so through a function that records each time it is opened, and the person sees each of these records. The copy is deleted when the time ends or when the person stops sharing.
Only our own code. The website’s security policy lets only its own scripts run on its pages, and the code libraries the app uses to read statements are served from our own site rather than from a public content delivery network. Otherwise, a script from anywhere else could reach a person’s key while their data is open. The one exception is the page where people in the US link their bank, bank.steerspend.com, which loads Stripe’s script: it is a separate address, so that script can’t reach anything the app keeps (section 3).
Plan status can only be written by our server-side function, which checks Stripe’s signature on every message.
Statement files are read on the person’s device and are not uploaded. Card details are handled only by Stripe, and the website is served over HTTPS.
Before a payment description is sent to the AI, the app takes out the person’s own name, other people’s names (each replaced by a code made with a key kept in the person’s account, so the code cannot be turned back into the name), account numbers, sort codes, card numbers, references, email addresses, phone numbers, postcodes and dates. Our server function then removes number patterns again and sends no name, email address or account ID. It keeps nothing it receives, and it accepts requests only from signed-in people with a running plan who haven’t switched the AI off, for up to 1,000 payees each a day. The Anthropic key is kept as a secret in Supabase, never in the web app, and the anonymiser is tested before every release.
Secret keys for outside services, such as Stripe, are never put in web pages, saved in source code or shared in messaging apps. If we suspect a secret has been exposed, we replace it the same day.
6. Monitoring and updates
We check the company inbox on working days and subscribe to security and status alerts from the suppliers listed in section 7, so that warnings are dealt with promptly. Before each release, we review the code for exposed secrets, data stored somewhere it should not be, anything that could let one person see another person’s data, and anything that could save information unencrypted or let code from another site run. Before release, the encryption was reviewed twice by an AI code reviewer working separately from the code’s author, and the problems found were fixed. This review is not a substitute for an independent penetration test (section 9). If a critical security fix is released for software our live service depends on, we apply it before the next release or within 14 days, whichever comes first.
7. Access control and incident response
We run SteerSpend through supplier accounts, including Supabase, Stripe, Hostinger, Resend and Anthropic. Each of these accounts must have its own password that is not used anywhere else, and must have two-step sign-in turned on. If anyone else ever needs access, they get their own login, which is removed on their last day.
The computer we use to run the service must have a strong password, lock itself when left idle and use full-disk encryption. Customer records are never kept on it, except a file a customer sends during a support request, which is deleted once the request is closed.
Before we take on a new supplier, we check its security and the access it needs, and then issue an updated version of this policy. In the US, bank connections go through Stripe Financial Connections, with read-only access (balances and payments only) under a restricted key used for nothing else, and Stripe is named in the privacy notice. Any other bank-data provider will have read-only access too, and we will name it in the privacy notice before bank connections are switched on in another country.
Anthropic was added as a supplier on 11 October 2026, with version 3 of this policy. It acts as our processor under its data processing terms, does not use what it receives through its API to train its AI, and deletes it within 30 days, unless it needs to keep it longer to enforce its usage policy or the law requires it. It receives only the anonymised descriptions described in section 5, never statement files or account details. Our Anthropic account has a monthly spend limit.
If there is a security incident, such as someone getting into data they should not see, a secret being exposed or an account being taken over, our first step is to contain it by replacing keys, locking accounts or switching features off. We then record what happened, when, and what was affected. If personal data is affected in a way that puts people at risk, we tell the people affected promptly. We also report it to the Information Commissioner’s Office without undue delay and, where possible, within 72 hours of finding out. Afterwards, we add what we changed to the risk register in section 8.
8. Risk register
These are the risks we watch as of 16 October 2026. The register applies to the whole company, and a new product that brings a new kind of information needs an update to this list before it launches.
One customer seeing another customer’s data. Mitigation: row-level security in the database returns only the rows that belong to the signed-in person, and each encrypted record opens only with its owner’s key. Any change to the database structure is checked against this before it goes live.
Someone with access to our database reading people’s payments. Risk: a supplier, an attacker or one of us could read financial records. Mitigation: these records are encrypted on the person’s device, and we never have the key. Support sees a person’s data only if they share it, for the time they choose, and each look is recorded for them to see.
Changed code on our website capturing keys. Risk: the app runs in the browser, so whoever controls the code our website serves could change it to read a person’s key while their data is open. Mitigation: the website’s security policy lets only our own scripts run, the code libraries are served from our own site, the hosting account has two-step sign-in, and each release is reviewed before it goes live.
A person losing access to their own data. Risk: the person loses their recovery code and has no device or passkey that can open the data. Mitigation: the code must be marked as saved before the app carries on, a new one can be made at any time while the data is open, a passkey can unlock the data on most phones and computers, and the unencrypted copy of an existing account is deleted only once the code is saved. If all of these are lost, the person can start again, keeping their account and plan.
Exposed secrets. Risk: a secret key left in public code or logs. Mitigation: only public keys are used in the web app, and secret keys are never put in web pages or source code. We check each release for exposed secrets before it goes live, and replace any secret we suspect has been exposed the same day.
Keeping data too long. Risk: holding statements or old payments for longer than needed. Mitigation: statement files are read on the person’s device and never sent to us. Purchases from more than about 12 weeks before the current pay period are removed each time the app opens and when a new statement is read. Deleting an account removes everything stored with it from the live database. We test that these deletions still work as part of every feature update.
Someone signing in as a customer. Mitigation: sign-in is handled by Supabase, and we never see passwords. Every account must also use a passkey, or an authenticator app code where a device cannot use a passkey. Customers can report suspected misuse to team@steerspend.com.
Loss or theft of the computer used to run the service. Mitigation: the computer must have a strong password, lock itself when idle and use full-disk encryption, and supplier accounts must have two-step sign-in. We check these controls at each scheduled review.
Customer data left in the support mailbox. Mitigation: support emails are deleted once the matter is resolved, and we review the mailbox at least every three months.
Personal details reaching the AI provider. Risk: a name or other identifier in an unusual payment description could be sent to Anthropic. These descriptions pass through our server unencrypted so that the AI can read them, although nothing is kept. Mitigation: descriptions are anonymised on the device and again on the server, requests carry no name, email address or account ID, and the person can switch the AI off and see exactly what was sent. Anthropic acts as our processor, does not train on the data and deletes it within 30 days. The anonymiser is tested before every release, and any gap found is fixed in the next release.
Misuse of the AI service. Risk: someone could use the AI function to run up costs, or a stolen key could be used elsewhere. Mitigation: only signed-in people with a running plan can use it, each for up to 1,000 payees a day. The Anthropic account has a monthly spend limit and two-step sign-in, and the key is kept only in Supabase and is replaced the same day if exposure is suspected.
Bank data from US bank connections. Risk: the bank function, or the page where people link their bank, could be used to read someone else’s bank data, or Stripe’s script could reach data it shouldn’t. Mitigation: the bank function checks the person’s sign-in, and that each Stripe session and account belongs to their own Stripe customer record. It uses a restricted key with Financial Connections access only, and asks Stripe only for balances and payments, never for account owners’ details, full account numbers or anything that could move money. Our server keeps only which accounts are linked; payments pass to the person’s device and are stored encrypted. Stripe’s script runs only on bank.steerspend.com, a separate address with its own security policy, and the session is handed to it by message, never in a web address. Linked accounts are unlinked at Stripe before an account is deleted, and Stripe stops reading them when a plan ends.
9. Keeping this policy up to date
We will revise this policy if our safeguards change, if we take on an important new supplier, or if a product starts handling a new kind of information. Each time, the version number and date change, and earlier versions are kept on file. To be open about where we stand: as of this version, we do not hold any external security certification, we have not yet commissioned an independent penetration test, and we do not run a round-the-clock security operations centre. If anyone asks about these, we will answer honestly.
Document control record
This policy applies to Algorythyme Ltd and to every software product the company runs.
| Entity name | Algorythyme Ltd |
|---|---|
| Company number | 16230652 |
| Registered office | Unit A, 82 James Carter Road, Mildenhall, IP28 7DE |
| Responsible | Algorythyme Ltd |
| Contact inbox | team@steerspend.com |
| Version and effective date | Version 8, 17 October 2026 |
The privacy notice is at /privacy.html.