Effective from 20 November 2026
1. Who we are
Hollow is an AI-native workspace suite at projecthollow.com, operated by Soulstack (“we”, “us”, “our”). Soulstack is the data controller for personal data processed through Hollow. Address for data-protection queries and legal correspondence: Office One, 1 Coldbath Square, Farringdon, London, EC1R 5HL, England. Email: privacy@projecthollow.com. See who runs Hollow.
2. What we collect
- Account data: your name, email address, password (stored only as a salted hash), profile and settings.
- Workspace content: everything you create or upload, such as notes, tasks, files, workbooks, slides, diagrams, budgets, contacts, calendar events, Yap messages, Cascade automations and Lumen monitors.
- AI requests: what you ask Hollow (in Hollow Studio, Hey Hollow voice or an app’s assistant panel), the workspace context sent with a request, and the answers.
- AI response reports: if you choose Report this response, we store the response text your browser sent (up to 10,000 characters), the app, the conversation or message it came from, a reason label and any note, a link to your account, and the steps administrators take to review it. Reports are private to Hollow administrators.
- Plan, billing and usage data: your plan, subscription status, invoices, your country and, for businesses, the VAT or tax number you give. Card details are collected and held by Stripe; we never see your full card number. We also record how much of each allowance you use (for example AI usage and storage) to apply your plan’s limits and show them to you. You can turn off new purchases for your account with the Allow purchases setting (Account, then Plan & billing), which is stored with your other settings.
- Sign-up and security data: sign-in times, session and device identifiers, two-step verification status, and the checks that keep sign-up and sign-in safe (Soulstack Verify, and Firebase App Check with Google reCAPTCHA, see section 6). To limit how many accounts are created, we keep scrambled (hashed) forms of the IP address and of a random device identifier your browser stores; we do not keep the raw values for this.
- Push notifications: when you enable notifications, Firebase Cloud Messaging tokens are stored so Hollow can deliver notifications to your signed-in device. You can change notification permissions in your device or browser settings.
- Location (Hollow Maps and Yap): your device’s location is read only when you ask for it, for example to centre the map on you or to share where you are, and it is not stored unless you save a place or send a location. Saved places, lists and their notes, a location you pin on a contact and a location you send in a Yap message are stored with your workspace content. Recent map searches are kept only in your browser.
- Voice and audio: when you dictate or send a Yap voice message, the audio is sent to Google Cloud Speech-to-Text (Chirp) to be transcribed. Yap huddles carry live audio and video between the people in the huddle; huddles are not recorded.
- Yap: messages, reactions, files, voice messages and locations you send, and who is in a huddle. Link previews and webhook avatars are fetched by Hollow’s own servers, so the linked site does not see your IP address. GIFs are different: your browser fetches them straight from KLIPY (see section 6), and a GIF message stores only KLIPY’s reference to the GIF and its title, not the image.
- Studio: documents you add to Knowledge are indexed so Hollow can search them, and the index is stored with your workspace content. Artifacts you publish, and share links you create, can be seen by anyone who has the link until you unpublish or remove them.
- Files: previews and downloads use links that expire after 15 minutes, and we record how much you download to apply your plan’s monthly download allowance. Files you share to Hollow Files from another app are kept in Files’ own browser storage on your device for up to 24 hours, and are uploaded only when you choose Upload files. Discard removes them.
- Files share links: if you share a file with a link, we store a random code, which file it opens, when you made the link, when it ends and the times it was opened in the last day. Anyone with the link can view or download that one file without an account. We do not ask visitors who they are and do not store who opened a link; a download through it counts towards your monthly download allowance, and we count requests per network address, using a salted hash that cannot be reversed once its daily salt is deleted, within two days, to stop abuse.
- Files and shares opened in an app: when you open a file in an installed Hollow app (for example with Open with on Windows), or share text or a link to Notes, Tasks or Studio, and the app cannot take it yet because it is still loading or you need to sign in, your browser keeps a temporary copy in that app’s own storage on your device: the file’s contents, name, type, size and last-modified time, or the shared title, text and link. The copy is not sent anywhere before the app imports it, and an app that is ready takes the file at once and keeps no copy. Importing works as if you had chosen the file yourself: what the app then saves (for example a note, a workbook or calendar events) is stored with your workspace content, while shared text and links, and files opened in Studio or Contacts, stay a draft or preview until you choose Save, Send or Import. Documents you import, such as Word files in Notes, are converted in your browser, not on a server, and pictures in a Word document are saved to your Notes storage like any image you add.
- Shared text and links: when you share text or a link to Notes, Tasks or Studio, the title, text and link are part of the address your device requests from Hollow’s web host, so they reach that host, which may record the address in its server logs (see section 8). No Hollow code reads them on a server. The page moves them out of the address straight away, and the app’s offline cache never stores them.
- Developer data: details of any developer keys you create and the requests made with them.
- Technical data: IP address, browser and device information and server logs, used to deliver Hollow, keep it secure and prevent abuse.
3. Why we use it, and our lawful basis
Purpose
Lawful basis (UK GDPR Article 6)
Creating and running your account and your plan
Contract
Providing the apps and storing your content
Contract
Running requests to Hollow with the model provider
Contract
Connecting an AI app you approve to your workspace and answering its tool calls (section 6A)
Contract
Taking payment, invoicing and handling refunds
Contract
Keeping tax and accounting records
Legal obligation
Applying plan limits and preventing abuse and fraud
Legitimate interests (see below)
Service emails: verification, receipts, renewal reminders, security alerts
Contract and legal obligation
Answering support and data protection requests
Legitimate interests and legal obligation
Handling content reports, complaints and appeals, and keeping records of them
Legal obligation (Online Safety Act 2023) and legitimate interests
Fixing problems and improving Hollow
Legitimate interests (see below)
Providing Hollow to a user who adds information about you (see section 3A)
Legitimate interests
Where we rely on legitimate interests, these are the interests:
- keeping Hollow, its users and their data secure, and preventing fraud, spam and abuse of accounts and sign-up;
- applying plan limits and fair use consistently, so the service stays available and affordable for everyone;
- finding and fixing faults and making Hollow work better, using technical and usage data, not your content;
- answering your questions and support requests, and keeping a record of them;
- keeping people safe by handling reports about content; and
- providing the service a user asked for when they store or share information about other people.
We weigh each interest against your rights and expectations, and you can object to any of them (section 9).
Hollow users sometimes store or share information about other people: contacts and addresses, calendar guests, names in notes and documents, people invited by email, and people who email us a report. We get this information from the user, not from you.
We use it only to provide Hollow to the user who added it (our legitimate interest in providing the service they asked for), and to send an invitation when a user asks us to. We do not contact these people for any other reason, use the information to train AI models, or sell it. It is kept for as long as the user keeps it. Because we usually cannot know who these people are or how to reach them, we publish this notice instead of telling each person (UK GDPR Article 14(5)(b)). If you think a Hollow user holds information about you, you can use the rights in section 9 by emailing privacy@projecthollow.com.
4. Where requests to Hollow are processed
At launch, AI routing is global on every plan. When you use Hollow, your request and its context are sent to the provider that runs the model, using that provider’s global service. Depending on capacity, the request may be processed in any region where the provider runs the model, including the United States and the European Union.
- Google Cloud Vertex AI: Gemini models, image and video generation, and open models (Gemma and Llama).
- Microsoft Azure AI Foundry: GPT, Grok, DeepSeek and Kimi K3.1 Pro models, and the Claude models that Microsoft hosts on Azure.
- Anthropic: Claude Fable 5.1. Anthropic hosts this model on its own infrastructure, through Microsoft Azure AI Foundry, so requests for it are processed by Anthropic and may take place outside Azure, including outside the region of the deployment.
- Amazon Bedrock: Kimi K3, and Hey Hollow voice.
- Google Cloud Speech-to-Text (Chirp): transcribing dictation and Yap voice messages, processed in the United States.
For Claude models, Anthropic acts as an independent data processor for prompts and outputs, whichever hosting option is used. Automatic safeguards can flag content for review by Anthropic’s trust and safety team, and Anthropic staff review flagged content only in exceptional cases, to investigate potential safety violations.
Each provider acts as our processor under its standard terms for business customers, which do not allow it to train its models on your requests. Your content is not used to train third-party AI models. We may offer processing kept within the EU or UK later, as an option for Hollow Business.
4A. Yap AI
Yap can write AI notes of a call. After a call, anyone who was in it can choose Summarise this call. Nothing is summarised, and nothing is sent to an AI model, unless a person who was in the call presses that button. Hollow never does it for you when a call ends.
- What is sent: when a participant presses Summarise this call, Hollow decrypts the chat messages sent in the conversation while the call was on (up to 60, including thread replies) on its servers and sends them to Google Cloud Vertex AI (Gemini, see section 4), with the @username of whoever sent each one (someone who has not chosen a username appears as Member 1, Member 2 and so on, never by name). Deleted messages and messages that disappear are left out. Messages from people another participant has blocked are still included, because the notes are shared with the whole conversation.
- Audio is never sent: only the text of the chat messages is. Huddles are not recorded, and call audio and video are never sent to an AI model.
- The result is shared: the notes are posted into the channel or conversation for everyone in it to read, marked as requested by the person who pressed the button. They are stored with your workspace content like any other Yap message.
- One request per call, paid for by the person who asks: a call can be summarised once, within 24 hours of it ending, and the request counts against the AI allowance of the person who pressed the button. So that the button works, we keep a short record of who was in the call and when it started and ended, with no messages in it, for 24 hours.
5. Where your data is stored
Your account and workspace content are stored with Google Cloud (Firebase). The main database is in a United States multi-region location, and parts of Hollow run in London and in the United States. Stripe processes payment data in the UK, the EU and the United States.
- Maps: map tiles and place search are served from Hollow’s own content delivery network at tiles.projecthollow.com, hosted on Amazon Web Services. Place search looks up Hollow’s own index there rather than a third-party search service, and these requests do not carry your account details.
- Canvases: live editing of Notes canvases runs on Hollow’s own sync service on Google Cloud in London, and canvases are saved with your workspace content.
- Web browsing: when Hollow or you browse the web in a browsing session, the pages are loaded by a separate browser on Hollow’s servers on Google Cloud in London, not by your device.
6. Who we share data with
- Google Cloud and Firebase: hosting, database, file storage, sign-in, functions, speech to text and push notifications (processor).
- AI model providers named in section 4 (processors).
- Stripe: payments, subscriptions and invoices. Stripe is our processor for most payments and an independent controller for its own fraud checks and legal duties. Where Stripe sells to you as merchant of record (customers in some countries), Stripe is the seller and its privacy policy applies to that sale.
- Amazon Web Services: sending service email (Amazon SES), running voice, carrying Yap huddles (Amazon IVS Real-Time, in Ireland, and, for HD screen sharing, the Amazon Chime SDK, in London) and serving maps (processor).
- KLIPY: GIFs in Yap. When you open the GIF tab, search for or browse GIFs, or a GIF appears on your screen in Yap, your browser sends the request straight to KLIPY, not through Hollow. KLIPY receives what you type in the GIF search, which GIFs are requested, your IP address and your browser and device information. It does not receive your name, email address or Hollow account details, or the text of your messages. KLIPY is not our processor: it decides how it uses this information, under its own privacy policy.
- Soulstack Verify: our own check that a person, not a bot, is signing up or signing in.
- Google reCAPTCHA, through Firebase App Check: checks that a request to our servers comes from the real Hollow app and not a script. It runs in the Hollow apps, and on these pages of this website: Sign in, Create account (including confirming your email and checkout), Forgot password, Reset password and Pricing. It does not run on the other pages. Google receives your IP address and browser and device signals to run the check (processor), and may also use them under its privacy policy and terms.
- People you share with: content you share (a notebook, a board, a Yap channel, an artifact) is visible to those accounts. In Hollow Business, the organisation’s owners and admins can see members’ names, emails, seats and usage.
- Services you connect: Cascade connectors, MCP servers and GitHub accounts receive what you send them, under their own terms.
We do not sell your personal data, and we do not run advertising or ad tracking.
Hollow Search at search.projecthollow.com is a free public service that works without a Hollow account. It has its own privacy notice, which applies when you search there or when an app opens a search for you.
6A. Connected AI apps and the MCP server
You can connect an AI app that supports MCP, such as Claude, to your own workspace through our MCP server at mcp.projecthollow.com. This section covers that connection. It happens only when you start it and approve it.
- What the app receives: only what the permissions you chose when you approved the connection allow, and only what a tool returns when you ask the app to use it, for example the text of a note or the list of your tasks. It acts as you, on your own data and on what others have shared with you, so what others shared with you can reach the app when you ask it to read it.
- What the third party does with it: once a tool has returned something to the app, the company that runs the app is responsible for it. For example, Anthropic processes what Claude receives under its own terms and privacy policy, not ours. Check them before you connect, and do not connect an app whose terms you do not accept.
- What we store about a connection: a grant record (the app’s name, the permissions you approved, when you approved it, when it ends, and when and why it was revoked), the short-lived sign-in request while you approve it, and a separate access key for the app that lasts an hour at a time and is replaced as you use it. The key’s record also holds when it was last used and how many requests it has made. Where an app registers itself with us, we keep the name and return addresses it gave. We keep keys and refresh tokens only as scrambled (hashed) values.
- How long: see the table in section 8.
- How to disconnect: in Settings, Developer settings, under Connected apps, choose Disconnect. Its access ends within a minute, and you can connect it again later. Every connection also ends by itself after 90 days, and deleting your Hollow account removes your connections and their records.
7. International transfers
We are based in the United Kingdom. Some of the providers above process data outside the UK, including in the United States and the European Union. Where they do, we rely on UK adequacy regulations (including the UK Extension to the EU-US Data Privacy Framework where the provider is certified) or on the International Data Transfer Addendum to the EU Standard Contractual Clauses. KLIPY is different: your browser sends GIF requests to it directly, and KLIPY may store and use them outside the UK under its own privacy policy.
8. How long we keep it
Account data and workspace content
While your account exists. Items you delete go to the app’s Trash or Bin first where it has one (see the next row). When you delete your account, we wait 14 days, and you can cancel in that time, then we delete it all (section 9). Copies in our backups are overwritten within about 28 days.
Items in a Trash or Bin
Studio and Tasks remove an item 30 days after you delete it. Files removes it after 7 days on Seed and 30 days on a paid plan. Emptying the Trash removes it at once. Deleting your account removes what is in a Trash too.
Your data export
The zip stays in your account for 7 days, then it is deleted. You can delete it sooner.
AI requests and answers
Kept in your conversation until you delete it or your account. Model providers process requests only to answer them, under their terms for business customers, and do not keep them to train models. Reports of AI responses are in their own row below.
AI response reports
Deletion: closed 90 days after closing, and anything 12 months after it was filed, whichever is earlier. That is, 90 days after an administrator closes a report, or 12 months after you filed it if that comes first. A daily clean-up is set up to delete expired reports, and they are also cleared whenever someone files a report or an administrator opens the review queue, so a report can stay a little past its date until the next clean-up. Deleted with your account, and included in Export my data.
Connected AI apps (MCP)
The sign-in request is deleted a day after it expires, which is 10 minutes after it starts. A connection (grant) ends 90 days after you approve it, and its record is deleted a day after that. An app’s access key lasts an hour; it is deleted 30 days after it expires, or 7 days after it is replaced by a newer one. A registration an app made with us is deleted after 180 days without use. Revoking an app in Settings, Developer settings ends its access at once.
Push notification tokens
Kept until you turn notifications off in the app’s notification settings, which removes that browser’s token. Also removed when your account is deleted, and when Firebase reports a token as no longer valid, the next time a notification is sent to it.
Temporary copy of a file or share you open in an app
At most 15 minutes, in that app’s own browser storage on your device, within limits of 25 MB per file, 20 items and 100 MB in total. Deleted when it is imported, when you save or cancel a share, when you sign out, when you clear the site’s data, or when it is 15 minutes old (removed the next time you open that app).
Files share links
A link ends after the days you chose, 7 by default and 30 at most, or sooner when you stop sharing or when you move it to the Bin. Ended links are removed the next time you open Files. A link is also removed when its file is deleted forever and when you delete your account, and it is in Export my data without its code.
Files shared to Hollow Files
Up to 24 hours, in Files’ own browser storage on your device. Deleted when you upload or discard them, or after 24 hours (removed the next time you open Files).
Voice and audio
Audio from Hey Hollow and from dictation is processed live and not stored by Hollow; the text is kept only in the conversation or message you send. Yap voice messages are kept as messages, until deleted or until a disappearing message expires. Huddles are not recorded.
GIF searches in Yap
Not kept by Hollow. A GIF you send is kept as a message, which stores only KLIPY’s reference to the GIF and its title. Any record KLIPY keeps of the requests your browser sends it is covered by KLIPY’s own privacy policy.
Published artifacts and share links
Public until you unpublish them or remove the link. Deleting an artifact unpublishes it straight away; it stays in the Bin for 30 days, then it is deleted.
Other content reports, complaints and appeals
Reports about content other than AI responses, complaints and appeals are kept as long as needed to deal with them and to keep the records the Online Safety Act 2023 requires.
Billing records and invoices
Six years after the end of the tax year they relate to, as UK tax law requires. They stay after you delete your account: the checkout and top-up records in Hollow, which we remove once the period is up, and the customer record and invoices in Stripe (6 years).
A record that you deleted your account
The date, when you asked and how many lines were on your list, under a reference made from your account id and the time you asked, with no name or email address. That is pseudonymous: it points back to you only for someone who already holds your account id. It is kept for 365 days, so we can answer a question about it, then removed.
Emails about a deletion
The confirmation, the reminder, the receipt and any failure notice stay in our email queue, with your address and the plan you saw, for 30 days after each is queued, then are removed. The cancel link is never kept there.
Usage records (for plan limits)
Up to 400 days.
Sign-up checks (hashed IP and device counters)
Up to 30 days.
Sign-ups whose email was never confirmed
Up to 30 days, then the unconfirmed account may be deleted.
Server logs
Normally up to 30 days, longer only to investigate a security issue.
9. Your rights
Under UK GDPR you can ask us to:
- give you a copy of your personal data (access);
- correct it (rectification);
- delete it (erasure);
- limit how we use it (restriction);
- give it to you in a portable format (Account, then Privacy & data, then Export my data, then Request export): we build one zip in the background, email you when it is ready, and keep it for 7 days. It holds, for every app, Notes and canvases: Markdown notes in folders, canvases as JSON with their images; Tasks: CSV and JSON; Budgets: CSV and JSON; Contacts: vCard and CSV; Calendar: ICS; Maps: GeoJSON; Workbooks: CSV for each sheet and the original data; Slides: Markdown outline, JSON and media; Diagrams: JSON; Files: the files as you uploaded them; Studio: conversations as Markdown, generated files, project files, prompts, memories and settings; Yap: JSON and the files you sent; Monitoring, automations and repositories: JSON; Account and plan: JSON. It leaves out passwords and keys, other people’s content, search indexes and version history. If a part of your data cannot be read, the email and the file say which part;
- stop using it on the basis of legitimate interests (objection); and
- stop using it where you gave consent (withdraw that consent at any time).
Email privacy@projecthollow.com. We reply within one month.
Delete your account in Account, then Privacy & data, then Delete account. You see exactly what will be deleted, what passes to someone else and what is kept before you confirm. We then wait 14 days: your account works as normal, you can download an export, and you can cancel in Account or from the link in the email we send you (we also email you the day before). Our deletion job runs once a day, so on or within a day of that date we check that nothing stands in the way and lock your account, so you can no longer sign in, and from then the deletion cannot be stopped. If something does stand in the way at that point (a paid plan that is still live, a Business organisation you own, a shared notebook or repository with other people in it), we pause the deletion, email you which, and delete nothing; you put it right and ask again. If it happens after the lock (someone was added to a shared notebook or repository you own), we pause it, tell you, and finish by hand once that is put right; we never delete a notebook or repository that other people are in. Once your account is locked we delete your data from every app, including the AI response reports you filed, and email you a receipt.
- Shared spaces: a Tasks, Yap or Monitoring Space, a budget, a contact space or a map you own that other people are in passes to one of them, who keeps everything in it, including what you added. A Space you are only a member of, you leave: what you added stays, shown as a former member or without your name. We do not delete your account while you own a shared notebook or a shared repository that other people are in, because their work lives in it. We tell you which, and you can stop sharing or remove them first.
- Yap: the messages you sent are deleted in every conversation. In a direct message the other person keeps their own messages and sees “Message deleted” where yours were. The empty row of each message you sent stays, with your account id and no content, so conversations keep their shape.
- What we keep: this is the whole list. Billing records for 6 years after the end of the tax year, as UK tax law requires (checkouts, top-ups and the refund and charge entries of our audit log; Stripe keeps the customer record and invoices, and we do not delete them), then they are removed. Records of any published page we took down for breaking the rules. The empty Yap rows above. A record that your account was deleted: the date, when you asked and how many lines were on your list, with no name or address, under a reference made from your account id (it is pseudonymous: it points back to you only for someone who already holds your account id), for 365 days. The emails about the deletion, including the receipt and any failure notice, which sit in our email queue with your address and the plan you saw, for 30 days after each is queued. Copies in our backups are overwritten within about 28 days.
10. Cookies and storage
Hollow uses no analytics, advertising or tracking cookies. We use only the cookies and browser storage needed to keep you signed in, to remember preferences such as your theme, to hold the random device identifier used for sign-up limits, to hold the temporary copies described in section 2, to keep an offline copy of an installed app’s pages and scripts, and to note on your device whether the app was first opened from the Microsoft Store, which is used only to show sign-in help and is never sent to us. Stripe’s checkout and billing pages set their own cookies for payment and fraud prevention, and Google reCAPTCHA (section 6) sets its own cookie on the pages and apps where it runs, to do its check.
11. Children
Hollow is not for anyone under 16, and we do not knowingly collect their data. If you think a child has given us personal data, contact us and we will delete it.
12. Changes to this policy
We will email you about important changes before they take effect. The date at the top shows when this version took effect, and what changed is listed here.
Changes to this page
- 20 November 2026: Updated the name of the assistant to Hollow throughout this notice.
- 18 November 2026: Removed the wording about requests for access to Hollow, in sections 3 and 8. Hollow no longer takes access requests now that sign-up is open to everyone.
- 15 November 2026: Added section 6A on connected AI apps and the MCP server at mcp.projecthollow.com: what an app receives, what we store about a connection, and how to disconnect. Section 8 now says how long a connection and its keys are kept. Section 3 gives the lawful basis for a connection (contract), and 6A says that deleting your account removes your connections.
- 6 November 2026: Removed the Hollow Mail wording. Hollow Mail is no longer offered, so the policy no longer describes connected mail accounts, or says that deleting your account does not delete mail Hollow holds.
- 6 November 2026: Section 4 now names Kimi K3.1 Pro among the models Microsoft Azure AI Foundry runs for the assistant.
- 31 October 2026: Section 4 now names Kimi K3 among the models Amazon Bedrock runs for the assistant, alongside Hey Hollow voice.
- 26 October 2026: Files share links: a link also ends when you move its file to the Bin, it keeps the times it was opened in the last day, and the network address counted to stop abuse is salted so it cannot be reversed once its daily salt is deleted, within two days.
- 26 October 2026: Files can now share a file with a link. Section 6 says what we store for a link and what visitors do and do not reveal, and section 8 says when a link ends and when it is removed.
- 26 October 2026: Export and deletion now work as section 9 describes: one zip of your data from every app, built in the background, and account deletion that waits 14 days, can be cancelled until it starts, and then locks your account at the next daily run, or pauses if something stands in the way. Also new: exactly what we keep after deletion and for how long, how Bin and Trash retention works, Google reCAPTCHA named where it runs, and this list of changes. Section 8 adds how long we keep a request for access to Hollow.
- 18 October 2026: Removed the model-training opt-in. Your content is not used to train AI models, so there is nothing to opt into.
- 13 October 2026: Named the Amazon Chime SDK in London for HD screen sharing in Yap calls.
13. Complaints
If you are unhappy with how we use your personal data, please complain to us first by emailing privacy@projecthollow.com or writing to Soulstack, Office One, 1 Coldbath Square, Farringdon, London, EC1R 5HL, England. We acknowledge your complaint within 30 days, look into it without undue delay, and tell you the outcome and the reasons for it (Data Protection Act 2018, section 164A).
If you are not happy with our answer, you can complain to the Information Commissioner’s Office (ICO):
Information Commissioner’s Office
Wycliffe House, Water Lane, Wilmslow, Cheshire SK9 5AF
Telephone: 0303 123 1113
ico.org.uk/make-a-complaint