A description, not the image
The image goes once to the reply model, which produces a short description. The rest of the chain only sees that description, anonymised like text.
A hosting customer often sends a screenshot instead of an error message. When image reading is enabled, 0seat describes it and takes it into account in the draft.
“It doesn’t work”, and an image attached. Without reading it, the assistant answers beside the point or asks for what the customer already sent.
The image goes once to the reply model, which produces a short description. The rest of the chain only sees that description, anonymised like text.
The error message or screen described enters the draft’s context, with the approved entries that relate to it.
A problem reported without an error message or screenshot triggers a request for details, with the instruction to mask any password.

It is sent to a provider configured not to retain data, for the time it takes to produce a description. Only the description is kept in the space.
Because a screenshot can contain data that text masking cannot see. Enabling it is your organisation’s decision, per space.
No. A document at the heart of the request leads to a handoff to a human.
Connect an inbox, approve the suggested knowledge base, and get your first prepared replies.
The draft arrives as a private note with the knowledge base entries behind it. The agent sends, edits or discards it. Below the threshold, no reply.
Every night, 0seat extracts question-and-answer pairs from your resolved conversations. Nothing is used until an agent has approved it.
Money, disputes, cancellations, a request for a person, a second follow-up, an answer missing from the base: an agent takes over, whatever the mode.