Privacy Policy
What Inklyst collects, who else sees it, and how long we keep it. Written from the code, in plain language, while the product is in invited beta.
Who we are
Inklyst is a research workspace operated by TwinBruhs LLC, a North Carolina limited liability company, at 4030 Wake Forest Road, STE 349, Raleigh, NC 27609, USA. When this policy says we or us, it means that company.
For the data your organization puts into Inklyst, your organization decides what goes in and why. We process it to run the product for you.
What this policy covers
It covers the Inklyst product, this website, and the email we send. It does not cover anything you reach by following a link out of Inklyst.
Inklyst is in invited beta. Surfaces change, and this page changes with them. The date at the top is the date of the last change.
What we collect
Three kinds of thing, and nothing else:
- Account details you type: your name, username, email address, and password. Passwords are hashed before they reach storage; we never see the original. If you sign in with Google, Microsoft, GitHub, or Apple instead, that service tells us your name and email address, and we keep the ID it uses for you so we know it's you next time. We keep no access to your account there, and we don't copy your profile photo.
- Content you put in a workspace: files you upload (audio, video, PDFs, captured web pages), the transcripts and text we extract from them, the citation anchors that tie an answer back to a passage, your chat threads and their answers, documents, boards, and the feed of what changed.
- Operational records the product needs to work: which organization and workspace a record belongs to, who did what and when, session and login events, and processing job state.
This site sets no advertising or analytics cookies. The product sets a session cookie so you stay signed in, and one more that remembers how you last signed in on this browser (a password, an emailed code, or which sign-in service), so the sign-in page can point you back to it. That cookie holds nothing else, lasts 30 days from your last sign-in, and stays after you sign out. There is no ad network and no data broker anywhere in this product.
How we use it
To run the product for you: to store and index your sources, transcribe media, summarize the external meetings we record for you, answer your questions against your own sources with citations, show your team what changed, keep your account secure, and email you about your account, your bookings, and what changed.
We do not sell your content. We do not share it for advertising. We do not use it to build or improve models of our own.
Our staff do not browse your content as a routine. Access happens for debugging when you ask us to look, and it is logged.
The tenancy boundary
Every record carries the organization it belongs to, and every read is filtered by that organization. Records that live inside a workspace carry the workspace too, and are filtered by both. There is no global query path, so a bug in one feature cannot pull another organization's files into view.
AI calls carry the same boundary. An answer is assembled from the sources in the workspace that asked the question, and from nothing else.
Who else processes your data
We use other companies to run parts of the product. Each one gets only what its job needs. This is the list as it stands for the beta.
| Provider | What it does for you | What it receives |
|---|---|---|
| Anthropic | Generates the answers in chat, and writes the summary after an external meeting we record for you | Your question, the passages we retrieved from your own sources, and the thread around it. To check an answer against a PDF or image you added, the original file, up to 2 MB. For a meeting summary, the transcript of that meeting, with each part marked by speaker number rather than by name |
| TypeSafe | Decides how to answer a chat question: from your sources, as a draft, or as a map of what your sources cover, and whether you asked us to look on the web | The text of your question, and what kind of answer came before it: facts, a draft or a map. Never your files or the passages we find |
| Voyage AI | Turns your sources into the search index that finds passages, and ranks the passages that answer your chat questions | Text and images extracted from the files in your workspace, and the text of your chat question when we look for the passages that answer it |
| Deepgram | Transcribes audio and video | A short-lived link to the media file you uploaded, or to the recording of an external meeting we captured for you |
| Datalab | Reads the text and the layout out of the PDFs you upload | The pages of the PDF, sent to be read |
| Firecrawl | Captures a web page you asked us to save | The web address you pointed us at, and the page it fetches back |
| Iframely | Turns a link you put on a board into a playable preview | The web address of the link, sent from our servers. The preview image loads from the linked site as soon as the card is shown. When someone plays the preview, their browser loads the player from Iframely's servers, like any embedded video |
| LiveKit | Runs calls inside the product and records them when you ask | Call audio and video, and the recording it writes back to our storage |
| Recall.ai | Joins an external meeting to capture it for you | Audio and video from the meeting you point it at, and, if you connect a calendar, access to that calendar's events and their attendees |
| Railway | Hosts the application, the database, the background workers, the search index, and the PDF export service | Everything the product keeps in its database, and everything passing through the application |
| Amazon Web Services (S3) | Stores the files themselves | Every file you upload, the media we transcribe, and every call recording, held at rest |
| Resend | Sends the four account emails we have: address verification codes, sign-in codes, organization invitations, and password resets. Also sends booking emails to you and the person who booked, and your summary of what changed, which you can turn off | Your email address and the contents of that message, including any one-time code or password reset link in it. A booking email also carries the guest's name, email address, and any note they left |
| Sentry | Collects crash and error reports so we can fix failures | A scrubbed copy of an application error: what failed and where in our code, with web addresses, keys and tokens blanked out, the request id, and a keyed hash of your organization's id. Never a request body, your documents, your name, or your email address |
Sentry receives the scrubbed error reports its row describes, from our servers and from the product in your browser. One vendor is wired in but not in the table: Stripe, for billing. It is not switched on for beta organizations. A beta organization gets its plan by a grant from us, with no Stripe subscription behind it, so Stripe receives nothing from a beta organization. We will add it to the table before that changes. Two other pieces look like vendors and are not: the search index and the PDF export service run on our own Railway infrastructure, inside the Railway row.
Hosting and providers are United States based today. The files themselves are stored in AWS's US West (Northern California) region. For everything else, region and routing follow each provider's own defaults. The service is offered from the United States. If you use it from somewhere else, your content and account data are transferred to and processed in the United States. During the invited beta we offer no separate data-transfer agreement, so do not put personal data that needs one into the service.
AI providers and model training
Every call the product makes to an AI provider is registered in our code with the provider, the credential, and the terms we sent it under, and every one of those registrations records terms that keep your content out of model training. That register is our own bookkeeping. It records what we asked for; it cannot prove what a provider did, so the two paragraphs below are the limit of what we will stand behind.
Two exceptions, both with the same date, because a blanket claim would be false. Audio sent to Deepgram, and the text and images sent to Voyage AI, went out before 2 September 2026 under each provider's own model-improvement default. On that date we made the Deepgram opt-out unconditional in the code and switched the Voyage opt-out on in their console. Everything sent after it carries an opt-out; nothing sent before it can be taken back. If that period matters to you, email us and we will tell you what it covered.
We have not signed zero-retention addenda with our AI providers yet. Until we do, each provider holds request data for its own published window. We are not going to describe that as zero retention while it is not.
How long we keep it
Workspace content stays until someone deletes it. Closing your account ends every one of your sessions immediately and takes you out of your organizations. It does not erase your user record: your name, username, email address and password hash stay in the database as a closed-account marker, because the product's history refers to them. Ask us and we will erase that record too.
Content you created inside a shared workspace stays with the workspace when you go, because your teammates are still working on it.
Deleting a workspace or an organization cascades: files, transcripts, citations, threads, and the search index built from them all go. Ask and we will bundle your content for export before it does.
The scheduled backup is switched on. Content you delete can survive in a backup until that backup is retired. There is no backup retention window to quote yet. Our security page says where that stands, and this section changes the day we quote one.
How we protect it
TLS in transit, hashed passwords, server-signed session cookies, role checks on the backend rather than in the interface, and the tenancy filter above. We are not SOC 2 certified and we have not had an external penetration test. Our security page is the honest version, kept current.
Your choices and your requests
AI features are switched on for every new organization, your personal workspace included. The owner of your organization, or someone they choose, can switch AI off for everyone, or just for direct messages, recorded calls and meetings, or certain kinds of content. In your personal workspace, that owner is you. Workspaces and files also have their own AI switch, and new ones start with it on. By default, on a recorded call the host can turn AI off for that call, and when someone from outside your organization is on a recording, AI stays off unless the host turns it on. Turning AI off stops new AI use; it does not delete your files or change who can open them.
In the product today you can see your account details, sign out of the browser you are in, and reset a password you have forgotten. Everything else is an email to us, and we do it by hand: a copy of your data, a correction to your details, signing out sessions on your other devices, or closing your account. We are building the self-serve versions, and this section changes the day they land.
One limit on closing an account: we cannot do it while you are the last owner of an organization or you still own a workspace, because that would strand your teammates. Hand those over and we will close it.
If your organization administers your account, we may route your request through them first, since the content belongs to their workspace.
Depending on where you live, you may have legal rights to access, correct, export, restrict or delete personal data about you, and to complain to a data-protection authority. To use them, email hello@inklyst.com. We answer within 30 days. If we need longer, or cannot do what you asked, we tell you and say why. The limits above still apply: closing an account leaves a closed-account marker until you ask us to erase it, and we cannot close it while you are the last owner of an organization.
Children
Inklyst is a workplace tool for adults. It is not for anyone under 16, and we do not knowingly collect anything from them.
Changes to this policy
When we change something that matters, we update the date at the top and email account owners. This policy is in force for the invited beta from the date at the top.
Contact us
Email hello@inklyst.com with anything about this policy, and we will answer.