Security & Privacy
DeskClone AI handles your customers' support conversations, your knowledge base, and your connected tools. Here is how that data is protected, including where a protection has limits.
Encrypted everywhere
All data is encrypted in transit with TLS and at rest - the database, file storage, and backups. Backups run automatically every day on a rolling window.
Scrubbed before the AI sees it
If a customer pastes a card number, SSN, or API key into chat, it is redacted before it is stored and before any AI model reads it - on every message, with no setting that turns it off. (If the scrubber itself ever fails, the message goes through and the failure is logged.) Outbound responses are scanned too, and anything shaped like a leaked secret is redacted from the stream.
Each customer is an island
Your agents, conversations, knowledge, and connections are scoped to your account on every query. Our adversarial isolation suite runs before every backend deploy made through our deploy pipeline. Knowledge files are private by default - only files you explicitly mark public can ever reach your chat widget. The exceptions are content you add on purpose for your customers, which starts out public: pages read from your own store or site when you connect it through our Shopify, Wix or WordPress app, facts you give Jarvis in the chat (documents you upload to Jarvis, and a website Jarvis reads for you, stay private until you choose what your agent can share), and imports through our developer API unless you mark them private. You can make any file private again at any time.
Connected accounts stay locked down
Credentials for the tools your agent uses - Shopify, Stripe, Zendesk, email inboxes, and the rest - are encrypted in the database and never exposed to the AI model. Our servers attach credentials only at the moment of the API call.
Customer chats can look, not touch
By default, website visitors' chat sessions use read-only tools. Write actions like refunds or order edits are unavailable to the model unless you enable them, and you can put server-enforced limits on them (maximum amounts, required fields) that prompt trickery cannot override. Your conversations are kept in your dashboard, answers drawn from your knowledge base show the files they came from, and you can export conversations as a CSV.
Prompt injection is expected, and handled
Chat messages are screened against known injection patterns - instruction overrides, role hijacks, prompt extraction. Agents that can move money (refunds and the like) are switched to blocking mode by default, and attempts are logged to a security event stream you can see.
Our commitments
Your conversations never train anyone's AI
Visitor conversations and your knowledge base are processed by OpenAI (or Anthropic, when our model settings select it) and indexed with AWS Bedrock, all under business API terms: your data is not used to train their models, and provider-side retention is limited to short abuse-monitoring windows. Every company that processes your data is listed in our Privacy Policy.
Where your data lives
All customer data is stored on AWS in the US region, encrypted at rest. Network access to the production database is limited to our application servers and a small set of named network ranges, and every connection is password-protected and TLS-enforced.
Deletion and retention
You can export or permanently delete an individual visitor's conversation data on request. Conversation data, including stored file uploads, is deleted automatically after a retention window: 365 days unless you ask us for a different one.
Responsible disclosure
Found a vulnerability? Report it to security@deskclone.ai. We acknowledge reports within two business days, keep you informed through triage and fix, and credit researchers who want credit.