Your data in your own bucket. Isolated by structure, not by promise.
Why it had to work this way
A product whose value is memory has to be uncompromising about privacy. The more 1Presence knows about you, the more it can do — and the more it would matter if that information leaked or got mixed up with someone else's. So we built the storage model with isolation as a structural property, not a policy.
Isolated by infrastructure
Every account gets its own private cloud storage bucket and its own dedicated agent server. There is no shared database where your notes sit next to someone else's. Access is enforced by IAM at the cloud level — not by application code that "checks the user ID before reading." Even if the application logic had a bug, the underlying cloud would refuse to serve another user's data.
- Per-user storage bucket. Plain markdown vault files live in cloud object storage scoped exclusively to you.
- Per-user agent pod. Your conversations run on a dedicated server that no other account can reach.
- Per-user service account. Authenticated via workload identity, with permissions to read only your prefix.
- Per-user memory. Knowledge graph and memory map live in isolated stores per account — and within your account you can keep a given agent's memory private to it, shared with the others, or fully walled off.
OAuth tokens you keep
When you connect Gmail or Notion, you authenticate to that service directly and we receive a token that lets 1Presence act on your behalf. Those tokens are stored in cloud secret management — encrypted at rest, never written to logs, never returned by the API.
You can revoke any connector with one tap. The token is destroyed immediately and 1Presence loses access on its next turn. You can also revoke from the service's side — Gmail, Notion, Slack all let you remove third-party apps any time.
Your conversations are not training data
Plain rule: we do not train on your conversations. We do not sell them. We do not feed them into anyone's improvement loop. The providers whose underlying models power your conversations have their own commitments — and we use commercial AI APIs whose contracts forbid training on customer content.
Aggregate, anonymised usage metrics — how many requests, how long sessions are, how often a feature is used — we do look at, so we can keep the product working. We do not look at the contents.
Confirmation before consequential action
1Presence will not send an email, post to social, or overwrite a vault file without explicit confirmation. The pattern is the same everywhere:
- Drafts are shown first. Email replies, social posts, calendar invites — always staged for your review before anything leaves your account.
- Vault overwrites require intent. 1Presence does not silently replace an existing file. Either you confirm the overwrite, or the new content lands in a new file.
- Access is granted once, on first use. The first time an agent reaches for something that leaves a mark on the world — an email out, a post, a file written to your Drive — it stops and asks. You decide what it may do (who it can write to, which channels, which folders), it remembers, and you can withdraw the grant any time from the Access tab. Read-only work just happens; the questions are saved for the moments that deserve them, and every grant is logged.
See it, edit it, take it with you
Privacy without portability is not privacy. Your vault is exportable at any time — every file, every memory entry, every shared document — as plain markdown and JSON you can read in any editor. If you ever leave 1Presence, you take your memory with you. Nothing is trapped inside the product.
What we will not do
- Sell your data, your conversations, or your metadata.
- Train models on your content.
- Give your data to advertisers, brokers, or anyone we do not have to.
- Hold your vault hostage if you cancel.
- Use dark patterns to keep you subscribed.
By design
Your conversations are not training data.
Your vault lives in a cloud bucket scoped to your account and enforced by IAM. OAuth tokens stay in encrypted secret storage. The models never train on what you say, and consequential actions wait for your consent the first time an agent attempts them. Disconnect anything, any time.
Start a conversation.
Free to try. No credit card. Just you and your agent.
Works on any device. Takes 60 seconds to start.