Six stages.Every AI under your rules.
SafeUs sits between your accounts and the AI that wants to use them. You sign in, connect what matters, and from then on every request is checked, the sensitive ones wait for you, and everything is recorded.
How it works
From your first sign-into a record you can prove.
No consultants, no agents to install on your laptop. Each stage is a screen you can see and a decision you keep.
Create your workspace.
Type your e-mail, enter the six-digit code we send, and say who SafeUs protects: you, your family or your organisation. There is no password to steal.
- Passwordless sign-in with a one-time code
- Optional two-step verification with an authenticator app
- Invite your team and give each person a role
Sign in · new workspace
No password stored
Connect what AI can reach.
Add your Google or Microsoft accounts, your websites and the AI agents you use. SafeUs maps who can reach what — reading metadata, never the content of your mail or files.
- Accounts, websites and agents in one map
- Provider tokens encrypted before they are stored
- Agents register with a manifest of the tools they use
Connected sources
Who can reach what — mapped
Every request passes the checkpoint.
When an agent asks for access, SafeUs checks who it is, what it wants, why, and which fields it needs. The answer is one of four: allow, limit, ask you, or stop.
- Only the fields a purpose needs are released
- Unknown or changed agents are held, not trusted
- Nothing is released if the record cannot be written
Access check · today
decision · reason · audit ref
You decide what matters.
Sending, deleting, paying or sharing waits for a person. You see the request in plain words, what SafeUs suggests, and you answer from your phone or the dashboard.
- Allow once, allow as suggested, or deny
- Requests expire if nobody answers
- Blocking an agent or Emergency Stop asks for your two-step code
Waiting for you
Sales copilot wants to send 40 e-mails
- Purpose
- Follow-up after the trade fair
- SafeUs suggests
- Allow for contacts met at the fair
expires if unanswered · recorded in the Black Box
The Black Box keeps the proof.
Every decision is written to an append-only, chained log: who asked, under which rule, who approved and what happened next. References and hashes — never content.
- Reconstruct any action, step by step
- Export records for your auditor or insurer
- History kept according to your plan
Black Box · timeline
sha256:0941a3f9…sha256:0942a3f9…sha256:0944a3f9…sha256:0951a3f9…Chained · append-only · exportable
When something looks wrong, act in one place.
Shield can watch, quarantine or block an agent. Emergency Stop shows you a preview, asks you to confirm, and ends with a report: what was done, what needs you, and what is outside SafeUs's reach.
- Suspicious signals put an agent under watch for seven days
- A Safety Check-up runs automatically every week
- Nothing irreversible happens without your confirmation
Emergency Stop · report
What SafeUs is not
Clear about what we do — and what we don't.
SafeUs does not scan your devices for malware. It controls what AI and connected services can do with your accounts.
We never store the content of your mail, files, chats or prompts — only decisions, field names and references.
SafeUs reduces risk on the access it controls and tells you plainly what is outside its reach.
Every status you see is backed by a record you can open and export. We never show a score we cannot prove.
Six stages.Start with the first.
Create your workspace with your e-mail — no password, no card. You can connect accounts and invite your team whenever you are ready.
Free plan · no card · cancel any time