01
Organisation boundary
Each org gets its own isolated database for knowledge, conversations and records, hosted in the region you choose.
How it works
Security is a product shape — not a checkbox afterthought
- 02
Members only
Only signed-in members of your organisation can manage agents and read conversations inside that boundary.
- 03
Scoped tools
Sub-agents only get the knowledge, forms and app connections you attach.
- 04
Secure embeds
Visitor sessions pass a bot check and get a short-lived token, and you can lock the widget to your own domains.
Capabilities
What 'secure by design' means day to day
The practical boundaries buyers and IT teams actually ask about.
Per-organisation isolation
Another customer's conversations and docs are not in your workspace — and yours are not in theirs.
Encrypted secrets
Your own AI provider keys and channel credentials, like Telegram bot tokens, are encrypted at rest with AES-256-GCM.
Knowledge stays yours
Uploaded docs and their search index live in your organisation's own database and are searched there.
Record privacy
Leads, tickets and feedback live with the organisation that collected them.
Least-privilege helpers
A product-lookup helper doesn't need email send access — and shouldn't get it.
Team roles and domain lock
Owner, admin and member roles for managing your team, plus a per-agent list of websites allowed to run your widget.
In practice
Questions security-minded buyers ask
Short answers without vendor jargon.
Is this shared multi-tenant soup?
No. Each organisation has its own database, so your conversations, knowledge and records are never stored alongside another customer's.
Can a visitor scrape our knowledge?
Visitors talk through the agent under the access rules you set — they don't get a raw dump of your library.
Who can change agents?
Signed-in members of your organisation. The public embed cannot reconfigure your agent.
What about connected apps?
App connections are authorised per organisation, and each helper gets only the apps and actions you pick.
FAQ
Questions
Is our data shared with other customers?
No. Each organisation has its own isolated database — conversations, knowledge and records never mix with another org.
Do visitors know they're talking to AI?
Yes. The widget shows an AI disclosure before the conversation starts, and each acknowledgement is recorded in that conversation's compliance log.
Is the website embed a public open API?
No. Visitors only get a short-lived session for one agent after a bot check. Add your domains to the agent's allowed websites and chat or voice sessions from anywhere else are refused.
Can I limit what an agent can touch?
Yes. Knowledge sources, Datastore and connected apps are attached intentionally — especially via focused sub-agents.
Where are the encryption keys stored?
In environment bindings, never in the database or code. Secrets are encrypted at rest with AES-256-GCM.
Where is our data hosted?
You pick a region when you create your organisation: Mumbai, Tokyo, Ireland, N. Virginia, Ohio or Oregon. Your organisation's database is created there.
Can we audit what the agent did?
Yes. Every conversation keeps a compliance log of the AI disclosure, inputs and outputs, requests for a human, helper activity and errors, viewable in your dashboard.
How do people sign in?
With email and password, a magic link or Google. Teammates join by email invitation.
Free plan · paid plans free for 7 days
Build your first customer agent.
No credit card. Questions? Write to support@embage.com.