Last updated October 11, 2026

Privacy notice

How AINA Technologies LLC (“we”, “us”), 1500 N Grant St Ste 45878, Denver, CO 80203, handles personal information in Shft. We are responsible for it as the “controller”. The short version: your agent works for you, your context and memory are kept by Shft, each step sends an AI model only what that step needs, we don't sell your information or use it for ads, and you can see, export or delete it.

What we collect

• Your account: your email address, name, time zone, sign-in sessions and the devices they're on.
• Getting in: if you join the waitlist, your email address and anything else you tell us, such as your name or what you'd hand off first. We use it to run the list and send invites, and we email you about your place on the list and your invite. If you join with someone's invite or link, we record who invited you and whether you joined, and we show them that their invite was used.
• Your agent's setup: its name, role, time zone, working hours, how much it may do on its own, and the rules and limits you set.
• Your work: what you ask, typed or spoken, the steps your agent takes, what it finds and makes, files, approvals, routines, reminders, receipts and briefs.
• Voice: when you send a voice note or talk with your agent, your speech is turned into text. We keep the text, not the recording.
• Memory: your agent learns from your work as it goes, without asking each time: your preferences, the people and projects you mention, and how you like things done. Each memory records where it came from, and you can review, correct or delete any of it in Settings. Something it picks up while reading outside content, such as a web page or someone else's email, waits for your review before it's used. It doesn't keep passwords, card numbers or keys, and it doesn't try to work out sensitive things about you, such as your health, religion, politics, sexuality or ethnic origin. It keeps something like that only if you tell it and ask it to remember.
• Morning briefs: if your agent sends you a brief, it's made from your calendar, the work that's waiting for you or has finished, email to your agent's address, and the weather for your city.
• A calendar address: if you paste your calendar's private address, we store it encrypted, never show it to AI models, and copy each event from yesterday to 60 days ahead (its title, time, place and whether it shows you as busy), refreshed every 15 minutes. We don't copy descriptions or guest lists. Reading it this way never changes your calendar, and removing the address deletes our copy.
• Apps you connect: if you connect an app such as Gmail, Google Calendar, Outlook, Slack, Notion or Linear, the connection's details and what your agent reads or writes there for a job, such as emails, events, messages and documents. If you connect Google Drive, only the files you pick. If you connect GitHub, the repositories you choose. Access tokens are encrypted or held by the service that runs the connection, and are never shown to AI models.
• Your agent's email: the addresses on shftmail.com (a personal one you claim, such as name@shftmail.com, and one for each agent, such as agent.name@shftmail.com), and the messages sent and received at them, including who they're with and what they say. If you email your agent a task, we keep that message and its attachments with the job. Only email from your own verified address can start a new task. Other email waits for you to look at it, or, if it's a reply to something your agent sent, comes back into that conversation. Either way your agent treats it as information, never as an instruction.
• Accounts your agent opens for you: when you ask it to sign up for a website, which site, the name and shftmail.com address it used, the password Shft made for it, and the emails the site sends back, such as a confirmation link. Shft keeps the password encrypted. You can reveal it after you confirm it's you again, or ask us to forget it.
• Websites and your agent's computer: the pages your agent visits, what it reads there and pictures of the pages it worked on, saved with the job so you can see what it saw. If your agent has its own computer, the files it keeps there between jobs and the websites you've signed in to in its browser, which stay on the computer while it sleeps and end after 30 days unused.
• Saved logins, passwords for accounts made for you, and your own AI keys: encrypted, and never shown to AI models. If you choose Keep me signed in for a website, its cookies are sealed and kept for no more than 30 days. If you connect OpenRouter in one click, OpenRouter makes a key for Shft, and we keep that key encrypted in the same way as a key you paste.
• The apps on your devices: on iPhone, iPad, Mac and Apple Watch, the device's name and type, a token for sending it notifications, and the approvals you give from it. With the Mac app, if you allow it: the apps your agent may use and pictures of them while it works, Mail and Notes if you turn them on, and the project folders you let it code in. If the Mac app crashes, it may send us a crash report with technical details only, never your content.
• Answers from other people: if someone answers a question your agent sent through a private link, their name and answer.
• Other AIs you connect: if you let another AI app use your Shft, by allowing it on Shft's own page or by making a personal token, the app's name as it gives it, where it sends you back, what you allowed it to do, and when it was last used. Each time it uses Shft we record which app, what it did and how many items it read or saved, never what they said. What it saves to your memory is marked with the app's name. We keep its tokens only in scrambled form.
• Agents you hand work to: where Shft offers it and you ask it to hand a job to another coding agent, the job, where it was sent, how it's going and what came back, such as a pull request link and a summary. Keys you add for those services are encrypted and never shown to AI models.
• Payments: your plan, your billing history, payment references from Stripe, a code Stripe gives for your card and a scrambled copy of your billing address, so each household gets one refund. Stripe and Link hold your card details; we never see your full card number.
• When you cancel or ask for a refund: your reason, if you give one, and whether you ticked “Email me if you fix this”. We use your reason to improve Shft, and we send that email only if you ticked the box.
• Usage: how much of your plan's included usage each job used, which we show you as a share of this 5-hour window and of this week, and what it cost us to run, to show you what's left and keep Shft fair and working.
• Security and service records: audit events, error reports, and basic request details such as IP address, browser and time. We may look at roughly where sign-ins come from, worked out from the IP address, to spot an account that's shared or used by someone else.
• Visits to our public pages (the home page, sign-in and these policies), not the app itself: pages viewed, where you came from, browser and device type, and whether you joined the waitlist. This uses no cookies or storage on your device, isn't linked to your account, and respects your browser's Do Not Track setting.

You need to give us an email address to have an account, and payment details to start a plan. Everything else is up to you, though some features need it to work.

How we use it

To provide Shft and do the work you ask for; to keep your agent's memory and history; to run your routines and briefs; to apply your plan's included usage and allowances and take payment; to check refunds; to send sign-in, invite, waitlist, service and (if you asked) comeback emails and notifications; to keep accounts and Shft secure and prevent misuse; to enforce our terms; to fix problems and keep Shft reliable; and to meet legal obligations. We don't sell your personal information, we don't share it for advertising, and we don't use your content to train AI models.

If you're in the European Economic Area (EEA), the UK or Switzerland, this is the legal basis for each use:

• Your account and the work you ask for: signing you in, your agent's work, memory, briefs, routines, connections, voice, your agent's email, notifications and the apps. Basis: our contract with you.
• Plans and payments: taking payment, applying your plan's included usage, pausing, canceling and refunds. Basis: our contract with you. Keeping payment and tax records: our legal obligations, or our legitimate interest in meeting the laws that apply to us.
• Other AIs you connect, agents you hand work to, and your own AI keys: sending them what you choose, and recording each use. Basis: our contract with you.
• One refund per household: the codes we compare, what we keep after you delete your account, and the record of what you agreed to. Basis: our legitimate interests in preventing misuse of refunds and in being able to show what you agreed to.
• Security: keeping accounts and Shft secure, preventing misuse, checking roughly where sign-ins come from, and error and crash reports. Basis: our legitimate interests.
• The waitlist and invites: Basis: the steps you ask us to take before you have an account, and our legitimate interest in running the list and invites.
• Visits to our public pages: counting them. Basis: our legitimate interests.
• Information about other people: what you give your agent or it receives while working for you, such as an email or an answer to its question. Basis: your and our legitimate interests in doing the work you ask for.
• Sensitive information you ask your agent to remember, such as something about your health. Basis: your explicit consent, which you can withdraw by deleting that memory.
• Legal claims and requests from authorities: Basis: our legal obligations, or our legitimate interest in protecting our rights and meeting the laws that apply to us.
• Comeback emails: the email we send if you ticked “Email me if you fix this”. Basis: your consent, which you can withdraw at any time by writing to us or with the link in the email.
• Anything else we ask your consent for. You can withdraw it at any time, which doesn't affect what we did before.

You can object to any use that relies on our legitimate interests.

Who at Shft sees it

People at Shft don't read your conversations, files, email or memory, except when you ask us to (for example, to sort out a problem), when it's needed to investigate misuse or a security problem, or when the law requires it. To run Shft, we see account and billing details and records of use, such as how much of your included usage you've used, and a person at Shft looks at the account and billing details involved when a refund is flagged for review.

Who else receives it

Other companies run parts of Shft for us. Each receives only what its part needs, and the one that does depends on the feature and the model. Requests may be processed in the US and other countries.

What the AI companies keep. OpenAI, Anthropic, Google, xAI and the hosts of open models process the requests we send them. Under their terms for API use, none of them trains its models on those requests. Some keep requests for a short time to check for abuse, up to 30 days at OpenAI and Anthropic, and for the period their own terms set elsewhere. Not every route is a no-retention route. Two rules always hold. Content from your Google Drive goes only to routes that keep nothing. Models from labs in China, such as DeepSeek and GLM, run only on hosts in the US that keep nothing, never on the maker's own service.

• AI models, through OpenRouter. Each step of a job goes to one AI model with only what that step needs, such as your request, relevant memories and tool results. Many steps go through OpenRouter (US), a gateway that passes each request to a company hosting the chosen model. When a request needs it, as the two rules above do, we ask OpenRouter to use only hosts that, under its terms, don't keep requests and don't train on them, and we don't send that request when no such host is available. Other requests may go to a host that keeps them for a short time to check for abuse, as that host's terms say. These hosts include Amazon Web Services (Bedrock), Google Cloud (Vertex AI), Microsoft Azure, Anthropic, OpenAI, xAI, Mistral AI, Groq and Cerebras, and for some open models other hosts such as Together AI, DeepInfra, Parasail and Baseten. OpenRouter keeps details of each request, such as its size and timing, but not what it says. Where a host needs a user ID, we send a scrambled one, never your name or email address.
• OpenAI, Anthropic, Google and xAI, directly. Some steps go straight to OpenAI, Anthropic, Google or xAI (all US), for example for GPT, Claude, Gemini or Grok models, or when a model's host through OpenRouter is unavailable. OpenAI and Anthropic may keep requests and responses for up to 30 days to check for abuse, and longer if a request is flagged for breaking their rules or the law requires it. Google and xAI keep them for the period their terms set. None of them uses them to train its models.
• Microsoft Azure. Microsoft runs some models and services for us on our own Azure resources, in Sweden, the US and other regions: some models from OpenAI, xAI and Microsoft, making and checking pictures, speech, and the search that finds your agent's memories. Microsoft may keep requests and responses for up to 30 days to check for abuse, and doesn't use them to train models.
• Cloudflare Workers AI. We don't send requests to it today. We'll update this notice before we do.
• Web search. Search words go to Brave Search, or to Microsoft's Bing through Azure, and a step may use OpenAI's or Anthropic's own search. We don't add your name, email address or account details to a search, but the words can contain what you include. Brave keeps the words of each search for up to 90 days. Bing searches fall under Microsoft's Bing terms, outside our other Azure terms, and may be processed in the US.
• Pictures. What you describe, and any picture you ask to change, go to Black Forest Labs' FLUX models on Microsoft Azure, and Azure AI Content Safety checks them. If Azure is unavailable, another picture model may make it through OpenRouter.
• Speech. Our own voice service, on our own server in the US, turns speech into text and text into speech. The audio is processed there. When it's unavailable, Microsoft Azure or OpenAI does it instead. If our voices are unavailable, your browser's own voice may read replies aloud, and some browsers do that through their maker's servers.
• Weather and website icons. For a brief, only your city's name goes to a weather service (wttr.in). When an answer lists its sources, our servers fetch each site's small icon, so the site sees a request from Shft, not from your device.
• Connections. To connect apps such as Gmail, Google Calendar, Slack and Notion, we use a connections provider, Composio (US). You sign in on the app's own screen, which may show the provider's name. The provider holds that connection's access and passes your agent's requests and the app's replies between them. It knows you by a code we make, not your name or email address. Google handles the Drive files you pick, and GitHub the repositories you choose.
• Email. Resend (US) sends and receives email for us: our sign-in, invite, notice and comeback emails, and your agent's email on shftmail.com, including what people send to your agent's addresses. Resend stores its data in the US, and keeps email and logs for 30 days on its standard plans, as its security page says. Your agent works from our own copy, which we keep as described under How long we keep it.
• Notifications. If you turn them on, Apple's push service delivers notifications to your Apple devices, and your browser's push service (such as Google, Apple, Mozilla or Microsoft) delivers web notifications, which are encrypted on the way. A notification may show a short preview, such as the name of a task.
• Hosting. Render runs the web app and API, and Supabase holds the database, both on the US East Coast. Contabo hosts our own server, also on the US East Coast, which runs your agent's browser, our voice service, our test builds and, when E2B isn't used, the worker that runs your agent's code. Speech audio, the pages your agent's browser works on, and the code and files of runs that don't use E2B, are processed on that server. Code runs there have no internet, and the run's files are deleted when it ends. Cloudflare delivers the Mac app and its updates, so it sees the address of a Mac that downloads them.
• Bot checks. When you sign in or join the waitlist on the web, Cloudflare Turnstile checks that a person is asking, not a script. For that check it sees your IP address and some details of your browser. It doesn't see what you type, and we don't use it for anything else.
• Code and your agent's computer. Code your agent runs, and its own computer if it has one, run on E2B (FoundryLabs, Inc., US). E2B keeps the computer while it sleeps, including its files and the websites you're signed in to in its browser. Files stay until you delete the agent or your account, and a sign-in you don't use for 30 days ends. Sign-ins to banks or government sites are never kept, on the computer or anywhere else: you sign in yourself for that task, and you're signed out when it ends. If we add backups of its files, they'll be kept with Cloudflare R2, and we'll update this notice first. The sandboxes that run code for your agent, including Python for number work, have no internet access and stop after ten minutes. Your agent's own computer, once you have one, can browse the web the way your agent's browser does. When Anthropic's code sandbox is used, the code and files that run needs go to Anthropic. Code and files can contain personal information.
• A proxy network. If a website blocks Shft's browser, Shft may try the page once more through a proxy network provider, which sends the request out from an ordinary home internet address. The provider sees which sites are visited on that retry, and how much data passes. We don't tell it who you are, and we don't use it for banks or government sites. We'll name the provider here once we've signed with one.
• Payments. Plans are paid through Stripe, using its Managed Payments and Link. Stripe is the merchant of record for indirect tax: it adds sales tax, VAT or GST where it applies. Stripe (US and Ireland) and Link handle your payment details, tax and receipts under their own privacy policies. We receive your plan, payment status and amounts, a code for your card and a scrambled copy of your billing address, not your full card number.
• Visitor counts. PostHog (US) counts visits to our public pages. Each visit reaches PostHog through our server, without your IP address.
• People you work with. Other people's agents can ask only what you've allowed. For scheduling that means free times, never your event titles, notes, guests or memory. If you share your agent, people with the link see how it works (its instructions, role and the skills you choose), never its memory, work, connections or the people it knows.
• Legal and safety. We may disclose information when the law requires it, or to protect the rights, property or safety of our users, the public or us. If our business is merged, acquired or sold, information may pass to the new owner under this notice.

Where you choose to send it. These services don't work for us. You choose to send them your information, and each one handles it under its own terms and privacy policy, not this notice.

• Your own AI key. If you add one, the work that uses it goes to the company you chose, under your agreement with it. On every plan, when your included usage runs out, Shft can carry on with your key on the same model: the first time, it asks you, and after that it follows your choice in Settings. Your provider's terms decide whether it keeps what you send or trains on it, so check its settings.
• OpenRouter, if you connect it. When you connect OpenRouter in one click, you approve it on OpenRouter's own page and OpenRouter gives Shft a key for your OpenRouter account. Work on that key goes through your account, under your agreement with OpenRouter. If you disconnect it, we delete our copy of the key and keep only which service it was for and its last four characters. The key itself works until you revoke it on OpenRouter.
• Other AIs you connect. You can let another AI app read your memory, files and tasks, save notes to your memory, or start tasks, and you choose which of these it may do. It never gets your passwords, saved logins, the websites you're signed in to, your agent's email, or memories waiting for your review, and we hold back anything that looks like a password, card number or key. Anything it reads is then handled by that app's provider under its own terms. A task it starts asks you in Shft for anything that needs your go-ahead. You can disconnect it at any time in Settings, and it stops working with your Shft at once.
• Agents you hand work to. Where Shft offers it, you can ask it to hand a job to another coding agent, either through GitHub, where Shft writes the issue or comment that asks for it as you, or through that agent's own service, using your own key. Shft asks you first. The job's description and the repository or files you choose then go to that service, and it handles them under its own terms. What it sends back is treated as information, not instructions.

Google user data

This covers files from Google Drive, and email and calendar events from Gmail and Google Calendar when you connect them through our connections provider.

• We use it only to do what you ask, in features you can see, such as reading a file for a report or finding an email.
• We don't sell it, use it for ads, or use it to train AI models.
• No person at Shft reads it, unless you ask us to for a specific message or file, it's needed for security or to investigate misuse, or the law requires it.
• Google Drive files go only to AI hosts that don't keep requests. They never go into your agent's memory or a web search, and we don't store their content. What your agent writes from them, such as a report, is your work and is kept like the rest.

Shft's use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements.

People who don't use Shft

Someone who uses Shft may have their agent, an AI assistant, email you from its shftmail.com address, ask you a question through a private link, or find a time to meet. Emails from its shftmail.com address say it's an AI agent and who it works for. If that person has connected their own email, their agent may write to you from that address, as them. What you send back is kept in that person's account so they and their agent can use it. Information about you may also be in what someone asks their agent to research, or in their connected email and calendar. If you don't want an agent on Shft to email you, or want to ask about information about you, write to hello@meetshft.com.

International transfers

We are a US company. Our database, web app, API and our own server are on the US East Coast, and some of our providers process information in other countries too. Where the law of the EEA, the UK or Switzerland requires a safeguard for information sent to a country it hasn't found to protect it adequately, we rely on these:

• The EU-US Data Privacy Framework, with its UK Extension and the Swiss-US Data Privacy Framework, for providers certified under them: Microsoft (Azure and Bing), Stripe, Cloudflare and Render. Resend is certified for the EEA and the UK.
• The European Commission's standard contractual clauses, with the UK's addendum and the changes Switzerland asks for, in each provider's data processing terms: OpenAI, Anthropic, OpenRouter, Supabase, Composio, E2B and PostHog, and Resend for Switzerland. Apple, which delivers notifications to Apple devices, uses standard contractual clauses for information from the EEA, the UK and Switzerland.
• OpenRouter's hosts. OpenRouter passes each request to the model's host under its own agreements with them.
• Search words. Brave Search gets the words of a search without your name or account details, and its data terms don't cover them.

When you send information to a service you choose, such as your own AI key's provider, an AI you connect or an agent you hand work to, that service's own terms cover it. You can ask us for a copy of these safeguards at hello@meetshft.com.

Cookies and storage

We use one necessary cookie to keep you signed in. Your browser also stores display choices, such as light or dark and your layout. We don't use advertising cookies, and visit counts on our public pages use no cookies.

Security

We use encryption in transit, secure HTTP-only session cookies, encrypted credentials that AI models never see, access limited to what each part of Shft needs, and an audit trail. Your agent treats outside content as information, not instructions. No system is completely secure, so we can't promise absolute security. If a breach puts your information at risk, we'll tell you and the authorities as the law requires. Never type passwords, sign-in codes or keys into a chat: use Saved logins, AI keys in Settings, or sign in yourself when your agent hands you its browser.

How long we keep it

• Your account, agent, memory and work: while your account exists. When a paid plan ends, whether you cancel, we refund you or a payment lapses, we keep your agent, memory and files for 90 days so you can come back, email you 7 days before we delete them, and then delete them. “Delete everything now” removes them at once, and a request to delete always comes before the 90 days.
• Sign-in links expire after 15 minutes. Sessions end after 30 days. Website sign-ins you ask Shft to keep (Keep me signed in) end after no more than 30 days.
• Your agent's own computer and its files: while the agent exists, including while the computer sleeps. Sign-ins in its browser end after 30 days unused, like Keep me signed in. Deleting the agent, or your account, deletes the computer, and when a paid plan ends it follows the 90 days above.
• Files you upload but don't send are deleted after a day.
• Other AIs you connect: access tokens last an hour and personal tokens last until you disconnect them. Disconnecting ends the connection at once. We keep a note of the connection and a record of each use in your activity log while your account exists.
• Your agent's email: 12 months, then it's deleted. You can delete any of it at any time, and deleting your account removes it. Resend keeps its own copy for 30 days on its standard plans. An address you stop using, because you release it, delete the agent or delete your account, is held for 90 days so mail isn't delivered to the wrong person, and is never given to anyone else within a year.
• Accounts made for you: the password, until you ask us to forget it, delete the agent or delete your account. Sign-ins to banks or government sites are never kept.
• Reply links stop working after 14 days, or the time your agent chose, and never later than 30 days. The answer stays with the job.
• Waitlist details: until you ask us to remove them. If you join, they stay with your account and are deleted with it.
• Crash reports from the Mac app are kept without your account and deleted after 90 days.
• Your reason for leaving and whether you ticked the comeback box: until we've sent the email or you ask us to stop, and for no more than 12 months.
• Deleting your account (Settings, Account), or choosing “Delete everything now”, removes your agents, memory, work, sessions and waitlist entry, and the connections we hold. It also ends a paid plan, so nothing more is charged. It doesn't refund a payment you've made, so take any refund you want first. A connected app may still list Shft, or our connections provider, until you remove it in that app's settings.
• After you delete your account, we keep a scrambled copy of your email address, a code Stripe gives for your card, a scrambled copy of your billing address, and whether you had a refund, so a refund stays one per household. We also keep the record of what you agreed to when you started a plan, including, in the EEA, the UK and Switzerland, your agreement that Shft starts straight away. We delete these after three years. We keep billing records as long as tax law requires.
• If you delete your account while some of your work is still finishing, we keep the email address we send the confirmation to until we've sent it, and never more than 7 days.
• Copies in backups are overwritten on their normal schedule. We keep payment and refund records as long as the law requires, and Stripe and Link keep theirs under their own policies. We may keep other records where the law requires or to resolve disputes.

Your rights and choices

You can download what we hold about you, or delete your account, in Settings, Account, and review or delete your agent's memories in Settings. The download is a file in a common, machine-readable format (JSON). If you're in the EEA, the UK or Switzerland, you have these rights:

• Access: to know whether we hold information about you, to get a copy of it, and to be told how we use it, where it came from and who receives it.
• Correction: to have information about you corrected or completed.
• Deletion: to have information about you deleted, unless we have to keep it, for example payment records.
• Portability: to get the information you gave us in a machine-readable file, or to have us send it to another company where we can.
• Objection and restriction: to object to any use that relies on our legitimate interests, to ask us to stop your agent's quiet learning or the morning brief, and to ask us to limit how we use your information while a question about it is settled.
• Withdrawing consent: where we rely on your consent, to withdraw it at any time. That doesn't affect what we did before.
• Complaining: to complain to us, and to a data protection authority: in the EEA, the authority in the country where you live or work or where you think the problem happened; in the UK, the Information Commissioner's Office; in Switzerland, the Federal Data Protection and Information Commissioner.

To use any of these, or to leave the waitlist, email hello@meetshft.com. We'll confirm it's you before acting, usually by asking you to write from the email address on your account. We'll answer within 30 days, and if a request is complex we'll tell you within those 30 days why we need longer, up to two more months. There's no charge. People elsewhere, for example in Brazil or a US state such as California or Colorado, may have similar rights, and they can ask in the same way. We won't treat you differently for using your rights.

Children

Shft is for people 18 and over. We don't knowingly collect information from children. If we learn that we have, we'll delete it.

Changes

We may update this notice. We'll post the new version here with its date, and tell you ahead of time about changes that matter.

Contact

AINA Technologies LLC, 1500 N Grant St Ste 45878, Denver, CO 80203 · hello@meetshft.com