Skip to content

Overview ​

Overdesk reads your support conversations so it can draft replies, which means it handles your customers' personal data. This section explains where that data goes, how it is protected, who on your team can reach it, how long Overdesk keeps it and how you remove it. It is written for the person who reviews Overdesk before your team relies on it.

The short version ​

  • Overdesk sends a reply itself only once your workspace turns sending on, which takes two keys: the Overdesk team enables it, then one of your admins turns it on. Until then every draft waits unsent in your help desk until a person on your team reads it and sends it. With sending on, only the replies your guardrails approve are sent, within the limits on Sending.
  • Drafting runs on your own Anthropic API key, so your conversations are processed under your own agreement with Anthropic.
  • Email addresses, phone numbers (international and North American), card numbers and other recognisable details are replaced with stand-ins before any conversation text reaches the model, then put back in the draft your agent reads. A phone number typed as one bare run of digits, or split by slashes, can reach the model as written, and so can a password written without a label or a username beside it, or one that is a short number, a word or a passphrase. Pictures and PDFs your customers attach are sent to the model as they are, unless you switch that off; see How customer data is handled.
  • Every query Overdesk runs on your workspace's behalf is limited to your workspace, and the few platform-wide jobs cannot show one workspace's content to another.
  • Help desk credentials, your Anthropic API key, the keys and tokens for the other services you connect, and webhook signing secrets are encrypted with AES-256-GCM using a key unique to your workspace.
  • An admin can erase one customer's conversations and the records built from them on request, and the owner can delete the whole workspace at any time.

Overdesk does not train models on your data. The Privacy Policy sets out that commitment in full.

What Overdesk writes to your help desk ​

Everything Overdesk writes into your help desk is for your team rather than your customer, such as internal notes, unsent drafts and tags. Three writes deserve a plain mention.

  • Escalation rules assign a conversation to a person or team in your help desk. The customer is not messaged. See Escalation.
  • Close stale conversations, which is off until your team schedules it, closes pending conversations nobody has touched for 90 days. Overdesk does not message the customer, but your help desk may send its own survey or notice when a conversation closes. See Closing stale conversations.
  • FreeScout gets an internal note and never an unsent draft reply. Overdesk cannot read which version of the API and Webhooks module you run, and a module older than 1.0.103 publishes a draft reply to the customer instead of holding it. A note never reaches your customer on any version. See Where drafts go.

In this section ​

Security questions the docs don't answer go to hello@overdesk.io.