Security & trust
Safe automation is the product, not an afterthought
We design for the failures that matter in customer service: wrong answers, unauthorised actions, data leaking between businesses, and AI that cannot be stopped.
Data protection
Strict tenant isolation
Every record belongs to one business workspace. Access is checked on every request and enforced in the database schema itself; automated tests attempt cross-business access on every release.
Encryption
Data is encrypted in transit with TLS. Integration credentials such as WhatsApp access tokens are encrypted at rest with AES-256-GCM and are never shown again after you save them.
Role-based access
Owners, admins, supervisors, support agents and read-only analysts each see and do only what their role allows — enforced on the server, not just hidden in the interface.
Secure sign-in
Passwords are hashed with Argon2id; sessions are revocable, HttpOnly and protected against cross-site request forgery; sign-in is rate-limited.
Data minimisation
AI models receive only what a conversation needs. Contact details are redacted from stored tool logs, and record lookups return summaries rather than full customer profiles.
Complete audit trail
Sign-ins, configuration changes, approvals, takeovers and AI actions are recorded in an append-only log that cannot be edited or deleted by the application.
AI safety controls
No invented facts
Answers must cite published knowledge or verified records from the same conversation. Replies that claim things no tool confirmed are blocked before they are sent.
Permissioned tools only
Agents can only use the tools you allow for their role, with validated inputs. They cannot run code, query databases directly or browse arbitrary websites.
Humans approve sensitive actions
Refunds and other high-impact actions are requests that wait for a person, and are re-checked against your records at approval time.
Prompt-injection defences
Customer messages, documents and records are treated as data, never instructions. Suspicious instructions in uploaded documents are flagged for review before publishing.
Kill switches
Pause any single agent, all AI handling in your workspace, or — for incidents — every automated reply across the platform. Paused conversations go straight to people.
Tested before release
Every configuration change runs a safety scenario suite first; publishing is blocked if any critical scenario fails.
Compliance and your data
We process personal data in line with the Nigeria Data Protection Act 2023. Our customers (the businesses using Orkesta) are the data controllers for their customers’ conversations; we act as their processor. Read our privacy policy for what we collect, how AI providers are used, retention, and your rights.
To report a security issue or request information for a vendor review, contact [privacy@your-domain.com] or use the contact form.
Questions from your security or compliance team?
We are happy to walk through our architecture, data flows and controls.