Grounded in the approved base, nothing else
The draft rests only on the approved entries provided, cited in the note. No procedure, price or timeline outside those entries.
Reply drafts
Prepared replies, with their sources, as private notes
On every incoming message, 0seat writes a draft grounded in approved entries and drops it as a private note; the entries it used are recorded in the console’s activity log. The agent sends, edits or discards it. The customer sees nothing before that.
The problem
An assistant that answers on its own eventually makes things up: a plausible procedure, a reassuring timeline, a wrong price. In a hosting support desk, an invented answer costs a customer.
How 0seat handles it
Hybrid search, with a threshold
Semantic similarity and full text, merged. Below the threshold, no draft is attempted: the conversation goes to a human.
A log per space
Every draft is logged with its decision, reason, entries consulted and cost.
What you control
- Each inbox’s mode: off or draft.
- The text that is sent: the draft is a proposal, the agent has the last word.
- The entries used, through approving and editing the base.
Honest limits
- Drafts per conversation and per hour and the daily draft budget are bounded by 0seat, not adjustable per space for now.
- When a problem is reported without an error message or screenshot, the draft asks for details instead of guessing.
- The draft promises no refund, goodwill gesture or timeline: those topics go to a human.
- Attached documents (PDFs, administrative papers) are not read.
Frequently asked questions
Does the draft cite its sources?
Yes, in the console’s activity log: each draft is tied there to the knowledge base entries that grounded it, with their scores. The private note itself holds only the suggested message, so a copy-paste never sends metadata to the customer.
What if no entry matches?
No answer is invented. The conversation goes to a human, with a suggested message the agent can edit.
Which language is the draft written in?
The draft follows the language of the customer’s message, in a formal register. Knowledge base search only works in French: a base in another language is not supported.
Can I review what the assistant proposed last week?
Yes, in the space’s log: decision, reason, entries consulted and cost of every draft.
Your team stays in charge. 0seat does the groundwork.
Connect an inbox, approve the suggested knowledge base, and get your first prepared replies.
See also
Knowledge base
Every night, 0seat extracts question-and-answer pairs from your resolved conversations. Nothing is used until an agent has approved it.
Screenshots
Images sent by the customer are described by the model and taken into account in the draft. Enabled per space; PDFs are never read.
Draft or autopilot
Off, draft or autopilot per inbox; autopilot opens per category once drafts are measured. Refused outright on payments inboxes and unofficial WhatsApp.