Tenant-isolated data
Every query is scoped to a tenant in the data layer rather than relying on each query to remember. One customer's agents cannot reach another's tickets, knowledge or reports.
Security and data protection
What we do with your data, and how it is protected.
Support tickets are among the most sensitive data a business holds: names, contact details, order histories, complaints, sometimes payment references. This page describes the controls built into the product today.
Separation and access
The two questions every support buyer asks first: can another customer see our data, and can our own staff see more than they should.
Role-based permissions
Privileged routes check a named permission, not merely that somebody is logged in. Roles and team access are configurable per workspace.
Argon2 password hashing
Agent passwords are hashed with Argon2, the current password-hashing standard, rather than a general-purpose hash.
No account enumeration
A failed login costs the same time whether the address exists or not. Equalising that closes a leak where identical error text still reveals real accounts through response timing.
Rate limiting
Public and authenticated endpoints are rate limited, including the intake routes that anyone on the internet can reach.
Encrypted provider credentials
API keys you supply for an AI provider are encrypted at rest rather than stored as readable text.
What the AI sees, and what is removed first
This is the part most support tools are vague about, so here it is precisely.
Redacted before the boundary
Email addresses, payment card numbers, phone numbers, provider credentials, and Aadhaar, PAN and GSTIN numbers are replaced before a request reaches an external model, then restored in the finished draft.
Grounded in your own knowledge
Answers are drafted from articles in your knowledge base, and only the articles actually cited are attached to the draft. Whether a draft counts as grounded is decided by the server, not reported by the model.
A person approves every reply
There is no action in the automation system that sends a customer-facing AI reply on its own. Automation handles tagging, priority, routing, assignment and status; a human decides what is said.
Stale sources are flagged
A cited article past its review date is marked on the draft, so an agent can see the answer rests on something nobody has checked recently.
Cached output is treated as personal data
A cached draft contains the restored personal values, so it is purged on erasure along with everything else. An erasure that skipped the cache would quietly leave the data behind.
No source means no answer
When nothing in your knowledge covers a request it goes to an agent rather than getting a guess. That is the designed outcome, not a failure.
The mechanism is described in more detail in how grounded AI works.
Retention, export and erasure
India's Digital Personal Data Protection Act, 2023 gives people rights over their data, and those rights are worth nothing if the software cannot act on them. Rezolo Desk has the operations built in rather than leaving them as a manual database job.
Data subject export
Returns a customer's record, their tickets and the messages they sent. Permission-gated, and the request itself is written to the audit log.
Data subject erasure
Anonymises the customer record, erases the message bodies they wrote, and purges cached AI output for them.
Cached data expires
Cached AI output, raw usage events and webhook delivery logs are deleted on a schedule rather than kept indefinitely.
Locked when you leave
A cancelled workspace is locked straight away: no sign-in, no changes and no new mail.
Immutable audit trail
Every mutation writes a tenant-scoped audit event. Personal detail is redacted inside the log entries themselves, so the audit trail is not a second copy of the data it describes.
Attachment controls
What file types and sizes may be attached is a workspace policy, enforced server-side rather than only in the browser.
Security questions
- Is customer data sent to an AI model?
- Ticket content is, with structured identifiers removed first. Email addresses, payment card numbers, phone numbers, provider credentials, and Aadhaar, PAN and GSTIN numbers are replaced before the request reaches an external model, then restored in the finished draft so the reply still reads naturally.
- Can the AI send a reply to our customer on its own?
- No. Rezolo Desk drafts and an agent approves. There is no action in the automation system that sends a customer-facing AI reply unattended. Automation changes ticket state, tags, priority, assignment and routing, which is the mechanical work rather than the answering.
- What happens when a customer asks for their data?
- There is a built-in export that returns that customer's record, their tickets and the messages they sent, and the request itself is written to the audit log. It is permission-gated, so only someone with settings access can run it.
- What happens when a customer asks to be erased?
- The customer record is anonymised, the message bodies they wrote are erased, and cached AI output for that customer is purged as well. That last part matters: a cached draft holds the restored personal data, so an erasure that skipped the cache would leave the data behind.
- How is support data separated between customers?
- Every query is scoped to a tenant, and the scoping is applied in the data layer rather than left to each individual query to remember. One customer's agents cannot reach another customer's tickets, knowledge base or reports.
Something here not covered, or need detail for a security review? Write to hello@rezolo.in and we will answer specifically. Our website privacy practices are in the privacy and DPDP notice.
Talk to us
Have a support queue you want to fix? Write to us.
Reach out at hello@rezolo.in. We will understand your requirements first, then guide you on the commercials.