Security

Your knowledge base and memory
stay in your hands.

Your company knowledge base runs on your own server; its memory stays on the computer doing the work, and your files and code are handled there too. The platform only handles sign-in, relays requests to the model and keeps usage and activity logs — it doesn’t store your files, knowledge base or memory. The table below lists exactly what lives on the platform.

Knowledge base

The two most valuable things
never live with us.

You choose where the knowledge base lives; memory always stays on the computer doing the work.

Just you

On your own computer

Documents stay on your own computer; it organises them and finds them by keyword. Works out of the box — no server needed.

Whole company

On your company’s own server

One shared library for everyone. Set it up in DooAny with one click — confirmed in person on the computer. Use an existing server, or have it buy one for you. The platform only authenticates and relays; it doesn’t store knowledge data.

Memory

Only on the computer doing the work

System addresses, ways of working, each person’s habits — everything it remembers is stored on that computer: never uploaded to the platform, never synced between computers. What it learns in a direct chat is used only for that person, and in a group only in that group; if it can’t tell who said something, it doesn’t store it. Open anyone in Contacts to see what it remembers about them.

01

Where things live, 
in one table.

The files and documents it works with stay with you. Two things to be upfront about: to generate answers, the relevant content is relayed through the platform to the model service; and the mobile app’s chats, meeting records and voiceprints are stored on the platform, so you can see your history on the phone and it can tell who’s speaking.

  • The platform stores neither your files nor your knowledge base.
  • Its memory stays on the computer doing the work — never uploaded to the platform, never synced between computers.
  • Model keys exist only on the server — never on computers or in installers.
02

Nothing that affects others: it just does it.
Anything that changes things: it asks first.

Research, reading files, drafting, and replying to whoever asked happen directly. Before changing your files, running commands or messaging others for you, it stops and shows you a confirmation card.

  • Default is “Ask me”. Switch to “Just do it” or “Don’t allow” as you see fit.
  • It can only read and write folders you allow — nothing else.
  • Paying, submitting mini programs for review and enabling tools it built can only be done by a person.
  • When you hand something off in DingTalk and aren’t at the computer, it won’t stop to ask about everything. Steps that need a person’s OK, like a release, it asks you about in DingTalk and waits for your reply.
03

Every step logged. 
Easy to trace.

Who asked, what was done, to what, whether it worked, why it failed, from which IP and when — all in the audit log.

  • Logs are isolated per company and visible only to your admins and owners.
  • Every entry carries a trace ID that matches server records.
04

Three roles. 
Nothing ambiguous.

Each company has exactly three roles — owner, admin, member — and they can’t be customised, so no one ends up with permissions nobody can explain.

  • Each company’s data is isolated; nothing is shared or queried across companies.
  • One person in two companies has two separate identities — departments, permissions and tasks don’t mix.
  • Submitting and approving documents are separate permissions — no one approves their own submission.
05

Keys and passwords 
exist only encrypted.

Model keys are the most valuable thing we hold: server-only, stored separately, ciphertext only. Passwords are stored only as one-way hashes.

  • Keys use AES-256-GCM . If ciphertext is tampered with, decryption fails outright instead of producing a wrong key.
  • Each ciphertext is bound to its service; moved anywhere else, it won’t decrypt.
  • In production, the service refuses to start without an encryption key.
  • Passwords use argon2 .
06

Approved sources only. 
No drafts.

For policy and business questions it only uses documents marked “Active”. Drafts, pending, expired or superseded documents are never used. Every answer cites its source.

  • When it writes to the knowledge base itself, it goes through de-duplication, classification and human review too.
  • When two documents contradict each other, both are paused until the owner resolves it.

FAQ

Will our data be used to train models?

No. We don’t store your files or knowledge base, and we don’t train models on your files, knowledge base or conversations. To answer a question, the relevant content is relayed through the platform to the model service.

Can we run the whole system on our own servers?

Work already happens on your own computers, and the knowledge base can run on your own server today. If you need a full deployment, mention it when you book a demo and we’ll assess it together.

Will it make decisions for people?

No. For spending, publishing, approvals and HR, it gathers information and evidence; a person takes the final step. The documents, sheets, code and conclusions it hands back note which sources it used and what it’s unsure of, so you can review them before use.

What happens when the balance runs out?

Calls stop — no overspending — and resume as soon as a top-up arrives. The admin console shows who used how much, and on which models.

Can data leak between two companies one person belongs to?

No. One person in different companies has separate identities; departments, permissions and tasks never mix.

What happens to our data if we stop?

Files and the knowledge base are already with you. Platform accounts, usage and logs are handled in order: stop running tasks, export data, confirm keep or delete, revoke connection credentials, then archive records.