Privacy Policy

• We do not sell your data, and we do not use your content to train any model — ours or anyone else's.
• One AI provider sees your content: OpenAI. Whatever a colleague reads or answers passes through their API.
• Memory is the product, so we keep it.A colleague's ledger is append-only by design — that is what makes it auditable, and it means individual memories cannot currently be erased one at a time.
• Deleting your workspace really deletes it. Personas, memory, tokens and stored files are purged, not flagged.
• Chat stays private.Web chat is deliberately excluded from the mechanism that lets colleagues see each other's memories.

Which of these you are changes who is answerable for the data, so it is worth ten seconds.

• You use Oolkin yourself — you sign in, create colleagues, feed them your work. We are the controller of that data and this policy governs it.
• You serve Oolkin colleagues to your own users through an API key bound to a subject namespace. Then the data belongs to your end users, you are the controller, and we process it on your instructions. Your own privacy policy governs what you tell them; this one governs what we do underneath. Write to privacy@oolkin.com for a data processing addendum before you launch, not after.

• Account information — your name, email address and workspace membership, handled by Clerk. We do not store your password; Clerk does not give it to us.
• What you create — colleagues and their definitions, the memories they form (judgements, facts and recorded reasoning), signed corrections, skills, board cards, chat messages, and any material you give a colleague to react to.
• Files you upload — portraits and documents attached to memories, stored as objects and reachable through the app.
• Derived copies — and this one is easy to miss. To make memory searchable we send text to OpenAI to be turned into a numeric embedding (text-embedding-3-small) and store that vector next to the original. Automated jobs also produce written summaries of your memories, which are themselves stored.
• Email you send a colleague — if you give a colleague an address, the messages sent to it are received, processed and stored, including sender and subject.
• Connected services — if you connect a GitHub repository, we store the repository name and receive the commit and pull-request events it sends us.
• Usage data — product analytics through PostHog, hosted in the European Union: pages viewed, features used, and similar events.
• Technical data — IP address, browser and device information, and server logs, collected automatically when you use the service.

This section exists because the ordinary privacy-policy vocabulary does not describe this product. Four behaviours are worth knowing before you put anything sensitive in.

• The ledger is append-only. Memories accumulate and are not rewritten. That is deliberate — a record you can silently edit is not evidence of anything — and it is why deletion works at the workspace level rather than memory by memory.
• Colleagues can see each other's memories, within your workspace. When a colleague records something, it can be linked to related memories held by other colleagues in the same workspace, so they can build on each other. This never crosses workspaces: every query is scoped to your workspace at the database level.
• Web chat is excluded from that. Chat is treated as private and is never offered to other colleagues. It is a deliberate exception, not an oversight.
• Some processing happens with nobody present. A scheduled job runs daily at 07:00 UTC and may read your memories to produce summaries. It runs on our infrastructure and sends content to OpenAI in the same way an interactive request does.

If you would not want a piece of information to persist in a record designed not to forget, do not give it to a colleague. That is the honest instruction, and it is a consequence of the design rather than a disclaimer about it.

Each of these processes data on our behalf, under their own terms, and only to the extent the service needs.

• Clerk — accounts, sign-in and workspace membership.
• Supabase — the Postgres database that holds everything you create, plus file storage and the live-update channel.
• OpenAI — the model that generates what a colleague says (gpt-5.4-mini) and the embeddings that make memory searchable. This is the only AI provider we send your content to.
• Cloudflare R2 — object storage for uploaded documents and artefacts.
• Resend — sending and receiving email on behalf of colleagues that have an address.
• Vercel — hosting, which means serving every request and its logs.
• PostHog — product analytics, on their European Union instance.

If you connect Oolkin to an outside AI tool over MCP, or mint an access token for one, that tool then reads whatever you have scoped it to. We cannot see or control what it does with what it reads. Tokens can be revoked at any time in settings.

Content a colleague reads or writes is sent to OpenAI's API to produce a response, and is then subject to OpenAI's own data policies rather than ours. Two things follow from that, and they are different from each other:

• Training. OpenAI states that content submitted through its API is not used to train its models by default.
• Retention.That is not the same as OpenAI holding nothing. Under its standard API terms, requests and responses may be retained for a limited period for abuse monitoring, and some request types are stored so a conversation can be continued across calls — which is how a colleague's tool use works today.

We say this plainly because the two get collapsed into “they don't keep anything”, and that would be a stronger claim than we can currently stand behind. If that changes — if we move to a zero-retention arrangement — this section changes with it and the date at the top will show when.

What we can commit to on our own behalf, today: we do not train models on your content, we do not use it to improve anyone else's colleagues, and we do not sell or rent it.

• While your workspace exists, its memory is retained. Retaining it is the service.
• Deleting your workspace, or leaving it as its last member, triggers a purge: your colleagues, every version of them, their memories and reviews, their access tokens and their stored files are deleted. This is a real delete, not a hidden flag.
• Individual memories cannot be deleted today. The ledger is append-only and no per-memory deletion exists in the product yet. If you need something specific removed, write to privacy@oolkin.com and we will do it by hand.
• Backups age out. Our database provider keeps point-in-time backups for operational recovery, so deleted data can persist in those for a short window after it has gone from the live service.

Depending on where you live, you may have the right to access, correct, delete, restrict or object to processing of your personal data, and to receive a copy of it. To exercise any of these, email privacy@oolkin.com. We will not charge you for it and we will not make you use a form.

Two honest notes. We do not yet have a self-service export button, so a copy request is fulfilled manually. And if you are an end user of a product built on Oolkin rather than an Oolkin account holder, ask that product's operator first — they decide what happens to your data and we act on their instructions.

We use cookies for two things: keeping you signed in, which is essential and cannot be turned off while you use the service, and product analytics through PostHog. You can block or clear cookies in your browser; blocking the essential ones will sign you out.

We do not run advertising cookies or third-party trackers.

• Workspaces are isolated in the database itself.The application connects as a role with no table permissions of its own and must adopt a workspace identity for any query to run at all — so a query that forgets to say which workspace it is for fails loudly instead of returning someone else's rows.
• Corrections are cryptographically signed.When you overrule a colleague's default, that rule is signed and chained, so the record of who changed what cannot be quietly rewritten.
• Access tokens are stored hashed, shown once when created, and revocable.
• Inbound email is authenticated before anything is filed, and senders can be restricted, because an email address is the easiest thing in the world to forge.

No system is perfect. If you find something, security@oolkin.com reaches us and we would rather hear it from you than from someone else.

Our analytics provider processes data in the European Union. Our other providers are United States companies and your data may be processed there or in other countries where they operate. Where transfers out of the European Economic Area or the United Kingdom are involved, they rely on the safeguards in those providers' own terms, such as standard contractual clauses.

Oolkin is not intended for anyone under 16, and we do not knowingly collect data from children. If you believe a child has given us personal data, write to privacy@oolkin.com and we will delete it.

If we change this policy we will post the new version here with a new date. If a change materially affects how we handle data you have already given us, we will tell account holders directly rather than relying on you noticing.

Questions, requests, or a disagreement with something on this page: privacy@oolkin.com.