Attributed, permanently
Every change, by anyone or anything, carries its author forever. Nothing can be edited anonymously or rewritten quietly.
Security
Your data has its own vault.
Every Memava customer gets a separate vault in its own container, with per-file access rules inside it, so your team's data is kept apart from every other customer's. Vaults run on the same host, and we say so plainly: the separation is containers, files and access rules, not separate machines.
Access
- Sign-in is by single-use email links that expire in 15 minutes, or a passkey (Face ID, Touch ID). There are no passwords to steal, reuse, phish, or write on a sticky note.
- Access is per file and per person: shared, read-only, or private. A private tree is exactly that: no admin, no AI, and no support request can open it.
- AI agents get their own credentials and only the permissions you explicitly grant. Revoking one takes a moment and touches nothing else.
Access
Everyone can edit deals/EDIT
Casey Park can read deals/harborview.mdREAD
Atlas (AI) can edit atlas/ onlyEDIT
private/jordan/ is Jordan's alone. No rule can open it.LOCKED
The same sentences bind people and AI. Agents never gain access from a wildcard.
Your data
Backed up daily
Encrypted in transit, backed up every day, and restorable per customer.
Never mined
Plain files you can export, every one you can read, any time, with one button. We never train models on your content, sell it, or read it outside a support request you initiate.
Leaving is easy
Minutes, and it takes every file you can read. A vendor you can leave is a vendor that has to behave.
AI, with adult supervision
An AI connected to your vault is a capable colleague, not an oracle. Its notes can be wrong, which is exactly why everything it writes is visible, signed, and reversible, and why there is an audit of what it read, not just what it wrote.
Memava does not make your AI right. It makes it accountable.
History of deals/harborview.md
What Atlas changed:
-cap rate holds at the broker number
+double-counted unit moves the cap rate 40 bps
... 12 unchanged lines ...
Reverting never erases anything: it adds a new version on top.
Questions security teams ask
Is my data stored with other customers' data?
No. Every customer gets a fully isolated instance that shares nothing with anyone else's: no pooled database, no shared index.
Can an AI agent read my private notes?
Only what its grants allow. Access rules are written per file and per agent; private folders are closed by construction, and an agent never inherits an everyone rule.
Do you train models on my content?
No. We never train on your content, sell it, or read it outside a support request you start.
Can I take my files with me?
Yes, in minutes, with one button: a zip of every note and file you can read, as plain files. Your own private notes come with it; other members' private notes never do. For the full history too, the vault can mirror itself to a git remote you own. Leaving is easy on purpose.
What happens if an agent writes something wrong?
Every change carries its author, human or AI, and any version can be restored. There is also an audit of what an agent read, not only what it wrote.
What stops instructions hidden in a note from steering an agent?
The server enforces every agent's grants on every call, whatever the agent was told, so a note can never widen what an agent may touch. Memava's tools tell agents that note bodies are data, not instructions. Switch a sensitive agent to Proposals only so a person accepts each change to a note before it lands, undo an agent's whole sitting in one step if something slips through, and set daily call and write quotas to cap the reach.
How are agent tokens handled?
Each agent gets its own token, and revoking one takes effect at once and touches nothing else. A client can send the token in an Authorization header instead of the URL, so it stays out of config files.
Do you offer SSO?
Not today. Sign-in is by single-use email link or passkey, so there are no passwords to steal.
Where is our data hosted?
In the United States, in our infrastructure provider's Ashburn, Virginia data center, as the privacy policy lists.
Working with your security team
This page covers the shared vault. For Memava Chief of Staff, the assistant that drafts your mail, the safety page sets out the rules it cannot break on its own, how we check them, and what we do not promise yet.
Our subprocessor list is short and lives in the privacy policy; a signed DPA comes with Office and Enterprise plans. Security questionnaires and disclosure reports both reach a human through the form, and vulnerability reports get answered personally and fast.
Evaluating any AI-connected notes tool, ours included? The MCP security checklist is the four-question audit we would run on ourselves.