AI customer support agent: answers drafted from your docs.
Repeat questions arrive with a drafted answer and its source. A person on your team sends it, and spends the day on the cases that need a person.
On requestBuilt for your company by the Unistaff team. Questions before the sale are answered today by the AI inbound sales manager.
Comparing tools first? The AI support assistant guide covers draft-first against auto-reply and what to ask a vendor.

In one paragraph
What does the AI customer support agent do?
The support role is an AI employee built on request for one company. It drafts the answer to a repeat customer question from your help pages, policies and past answers, in the language the customer wrote in, and hands anything your material does not cover to a person with the conversation attached. It has no send path: a person on your team sends every answer.
Next step
A scoping call about the questions you get, the material that answers them and who on your team sends. Price and timeline are agreed there, before anything is built.
What it does
Answers from your docs
Drafts the reply to a repeat question out of your help pages, policies and past answers, and names the source it used so an agent can check it.
Says when it does not know
When your material does not cover the question, the draft says so instead of inventing a policy, and the gap is shown to you.
Hands over the hard cases
Complaints, refunds, cancellations and anything outside the material go to a person with the conversation and the reason attached.
In the customer's language
Drafts in the language the customer wrote in, from the same approved material, so one set of docs serves every customer.
The path of one support question
Five steps from a customer’s question to an answer your team sent. The machine works steps two and three; a person owns the send and the hard cases.
- 01
Question arrives
Through a form on your site or by email, as agreed in the setup
- 02
Material checked
Your help pages, policies and past answers, nothing else
- 03
Draft or handover
A drafted answer with its source, or a handover with the reason
- 04
A person sends
Reads the draft, edits or sends it, answers the hard case
- 05
Gaps reviewed
Questions the material did not cover are listed for you
Never by itself
Sending an answer, issuing a refund or a cancellation, granting an exception to your policy, inventing a policy your material does not hold. Each of these waits for a person.
What exists today
- On request
Drafts from your docs
Your help pages and past answers become the material; every draft names its source and a person sends it.
- Live
Questions before the sale
Answered today on web chat and email by the AI inbound sales manager, from your knowledge folder.
- On request
Helpdesk connection
None ships today. A setup can build one against your helpdesk; until then questions arrive through a form or email.
- Planned
Telegram, WhatsApp, Instagram
On the roadmap for the same cabinet and the same handover rules.
How a setup starts
The support desk is built on request. It starts with a scoping call about the questions you get most, the help pages and policies that answer them, where questions arrive and who on your team sends. Scope and price are agreed before anything is built.
Then every repeat question arrives with a drafted answer, and the questions your material did not cover show you what to add.
Evidence · AI support
What it is built on
The pieces of the platform this role is built on, each marked with the site's three words: Live works in the product today, On request is built with you during a setup, Planned is on the roadmap.
Live
Questions before the sale are answered today
Pre-sale questions on your web chat and email are answered by the AI inbound sales manager, from your approved knowledge folder, with the AI disclosure in the first message and a handover to a person on complaints, emergencies and the word HUMAN.
On request
Answers after the sale are drafted and sent by a person
The support desk drafts from your own material and has no send path of its own. When the material does not cover the question, the draft says so instead of inventing a plausible policy.
On request
A helpdesk connection is built during the setup
Zendesk, Intercom, Freshdesk and the rest are connections built against your account during a setup. None ships out of the box, and the integrations page says so for each one.
The rules held in code behind every reply — AI disclosure, approvals, trust levels, pause and the stop button — are listed on the product page; the controls that exist and the ones that do not yet are on security.
Frequently asked questions
The support role is an AI employee built on request for one company. It drafts the answer to a repeat customer question from your help pages, policies and past answers, in the language the customer wrote in, and hands anything your material does not cover to a person with the conversation attached. It has no send path: a person on your team sends every answer.
On this desk a person on your team reads and sends every answer, so what reaches the customer is your team writing with a head start. Where a Unistaff employee talks to customers directly, as the inbound sales manager does, it says it is an AI assistant in the first message and confirms it when asked.
Your help center, PDFs and policies become the material it drafts from, and you approve that material before the first draft. How long a setup takes is agreed on the scoping call, together with the price.
It drafts the explanation of your policy. Refunds, cancellations, exceptions to a policy and anything that moves money stay with your team, and the platform has no payment path.
No helpdesk connector ships today. Questions reach the desk through a form on your site or by email; a connection to Zendesk, Intercom, Freshdesk or another helpdesk is scoped and built during a setup, and the integrations page carries each status.
The person who takes over gets the conversation so far, what the assistant drafted and the reason it stopped, so the customer does not have to start again.
Count the share of drafts that went out unedited, the time your team spent on repeat questions before and after, and the gaps found in your material. Resolution matters more than deflection: a question counts as answered only when the customer got the right answer.
The support desk is built on request and quoted after a scoping call. Scope and price are agreed before anything is built, and no list price is published for on-request desks. The free 30-day pilot applies to the AI sales manager.
Hand over the repeat questions
Bring your twenty most frequent questions and the pages that answer them to the scoping call. It settles where questions arrive, who sends and the price. The team replies within one business day.
Continue learning
Guide
AI support assistant guide: draft-first or auto-reply, what to ask a vendor
Blog
AI customer service: deflection vs. resolution, and how to count them
Live role
AI inbound sales manager: answers pre-sale questions on web chat and email
Compare
AI employee vs. chatbot vs. assistant
Blog
Customer support at scale without losing the human touch
Integrations