The chat with the AI agent is the place where you set tasks in words: “create the clients from this spreadsheet”, “merge these duplicates”, “prepare an invoice for this deal”. The chat lives right inside the product’s interface, so there is no need to go to another window and copy data there: the data and the task are in one place.
Why a chat inside the product
The product’s main idea is that the agent keeps the data. So there must be a simple way to explain to it what is needed. The simple way is the chat: you write in ordinary words, the agent works out which management commands this needs and executes them. No forms, no fields, no instructions about where to click. Next to the chat are the same data, so the result can be checked at once: you asked to merge the duplicates, and the queue refreshed before your eyes.
Connecting
The chat works through your Xedant Agent — the same one that gives access to the AI model. At installation two values are given: the agent’s address and its key. The names of these values are the familiar ones and match the Agent product itself, so they are easy to check when configuring.
If these values are not set, the chat is hidden — and the product says honestly why: “agent is not configured”. Everything else keeps working meanwhile: the client base, the timeline, the screens, request intake. Without the agent you simply cannot entrust changes in words — the data would have to be kept by management commands by hand, or the agent connected later. This is a conscious boundary: the product does not pretend it can do what it lacks.
What the agent can do
In essence — everything available for management, which is your whole build. In practice the most frequent asks are:
- Create and clean the base — upload clients from a spreadsheet, remove the repeats, bring phones and names to one form. See Contacts & Companies and Data Quality & Duplicates.
- Take in and route a request — create a lead, assign the owner, set the reaction deadline. See Leads.
- Set up the pipeline — create the stages, the transition conditions, the routing rules. See Deals & Pipeline.
- Prepare a quote and an invoice — assemble the lines, compute the price by the book, apply the discounts, issue. See Money: Quotes, Invoices, Payments & Contracts.
- Assemble a mailing — select the audience, check the consents, send, show the result. See Marketing.
- Review the duplicates — show the comparison and merge the disputed records by your decision.
- Answer questions about the data — “how many deals we closed this month”, “which clients have not bought in a while”, “which requests are overdue”. The agent looks into your data instead of inventing numbers.
Commands right from the screens
So you do not have to invent the wordings yourself, use the ready phrases. Next to the fitting place lies an “instruction for the agent”: press it — the phrase is already typed into the chat, only send it. For example, the duplicates queue holds “merge these duplicates”, the deal card holds “prepare a quote for this deal”, the intake console holds “sort out why this request was rejected”. If the chat is not connected, the same button copies a management command — you can pass it to the agent by hand.
Such hints sit almost everywhere: in cards, queues, add-on consoles. This is both training and speed: over time you start writing to the agent yourself, but the buttons are always at hand.
The chat panel
The chat opens from the menu on the left, where the Chats section lives. It shows the list of your conversations with the agent: you can return to a past one, continue it or start a new one. Messages arrive as the agent answers, and if it asks a question — you can answer right there without switching. The font size and the look come from your settings, so reading is comfortable.
Without the agent
If for some reason the agent is not connected, the product stays fully workable for watching: the screens, lists, cards, reports, request intake — everything is in place. Changes in that case are made by management commands: every “act through the agent” button can copy a ready command, which can be passed to the agent in another window or executed by hand. This is a fallback path, described honestly rather than hidden — but it is truly convenient when the agent is connected.
License
The license restricts only one thing: sending a message to the AI agent chat. Everything else — the work with data, the screens, request intake, management over the API, the add-ons — works without a license. This is a deliberate decision: first you make sure the product suits you, and only then decide about the license for the chat itself.
If the license term has expired, an attempt to send a message brings a clear notification offering to renew it — no mysterious refusals and no silently vanishing messages. The other work is not interrupted meanwhile.
What to expect from the model
The agent answers on the basis of your data: it does not invent numbers but looks into the base, and when something is missing it says so plainly. It can be wrong — like any assistant — so the important actions (merging records, bulk edits, sending a mailing) are arranged so that before execution you see what will come out, and after — what exactly happened. The check stays with the human, and that is not a shortcoming but a conscious order of work.
A subtlety about the model: you choose it in your own Xedant Agent, so both the quality of the answers and the cost of the model calls depend on your choice and settings — the product does not interfere and does not pass your data to anyone else. A stronger model means better answers and a bigger spend; a lighter one the other way round.
That is all you need to know for the work. If something stays unclear — look into the FAQ, and for a concrete screen’s description — into the documentation table of contents.
Next → Documentation table of contents
← Documentation table of contents · FAQ · Installation · Licensing