New AI Agent can now build your knowledge base, connect channels and invite agents — all by chat Try it now
Help homeProductPricingBlog
Start freeLog in

Security, permissions, and data boundaries

Understand workspace isolation, sensitive information, AI authorization, and high-risk action controls.

Updated 2026-07-17

YundaDesk isolates customers, conversations, knowledge, skills, memory, and channel configuration by workspace. Member roles further restrict what each person can view or perform.

Protect credentials#

Channel tokens, API keys, mailbox passwords, and store secrets must be submitted only through secure configuration fields in the product. Never place them in knowledge, Yuna conversations, customer messages, screenshots, or shared documents.

AI authorization#

  • AI can use only capabilities registered and ready for the current workspace.
  • Permission is checked again when a tool runs.
  • Refunds, price changes, and outbound messages require confirmation or approval.
  • The capability center describes readiness; it does not replace execution authorization.
  • When a tool, memory service, or retrieval system fails, the AI should degrade or refuse rather than invent success.

Knowledge and memory#

Use knowledge for business facts, AI skills for reusable procedures, customer memory for one customer's context, and Yuna memory for merchant-team preferences. Memory must not bypass knowledge review or hold unnecessary sensitive content.

Use individual accounts, assign least privilege, review login records, and revoke access promptly when a member leaves. Handle customer export, correction, and deletion through the governed product workflow and applicable law.

Didn't find your answer?

Start free and try the real product, or reach out — humans are as online as the AI.