Project AI assistant: setup, knowledge and hand-over to your team

SENSEVER

The official Sensever channel: guides and API documentation for Chat, Requests, Flow, Studio, bots and messaging channels.

Окууга 9 мүн

Project AI assistant: setup, knowledge and hand-over to your team

A project's AI assistant answers customers in the project's requests: from the website widget, the contact page and connected messengers. It works inside an ordinary request, so your team sees every answer and can step in at any moment. This article is a reference for the project's "Assistant" tile: every block, what it does and how the parts work together. For a step-by-step setup of a website chat, see AI chat for your website.

Where to find it and who can change it

Open Projects, choose the project and click the "Assistant" tile. The assistant, its knowledge and its data connectors may be changed by the project's owner and admins; agents can't change them, and the server refuses their requests. In the web app the project tiles, "Assistant" among them, are on the project owner's page; an admin's project page shows the "Request routing and team limits" and "Channel access" cards.

From top to bottom the tile has:

  1. the "Virtual assistant" switch;

  2. "Who answers customers": two tiles, Sensever model and Your server;

  3. for the Sensever model: the bot, "Bot name" and "Personality";

  4. "Knowledge base channels", "Files and tables", "Data connectors" and the "Test chat";

  5. "Save settings".

The switch, the choice of who answers and the bot are saved with "Save settings". The knowledge base channels have their own "Save" button. Files, tables and data connectors are saved right away, each in its own form.

Who answers customers

Sensever model. Our model answers from the materials you connect: knowledge base channels, your own files and tables, and data connectors. It reads only what is open both to the bot and to the customer who asks. Your agents stay at hand.

Your server. Sensever only forwards customer messages to your bot's webhook URL and passes your server's replies back; we generate nothing ourselves. Choose the "Bot that receives messages" from your own bots. Create a bot in chats with "Create → Bot" and add a webhook URL in its settings: bots without a webhook can't be selected. In this mode the knowledge blocks, data connectors and the test chat are hidden, because the answers come from your code. More: How a bot answers website visitors as a project.

Switching between the two tiles doesn't lose what you entered: each tile keeps its own fields until you save.

The bot and its name

With the Sensever model, the "Bot" list offers the project bot and, under "My bots", the bots you created yourself. If the project has no bot yet, the first item is "Create automatically": the bot is created when you save.

  • Bot name. A project bot gets the project's name when it is created. You can change it in "Bot name"; customers see this name on the assistant's messages.

  • Personality. Eva is warm and attentive, Leo is calm and concise, Leia is lively and friendly. The personality changes only the manner of replies, not the facts.

The name and personality belong to the bot, so they are the same in every project where this bot is selected.

Knowledge base

The assistant doesn't need training. It looks up the answer in your materials every time a question comes in, so an updated material is used in the next answer.

Knowledge base channels. The list shows the channels you own or manage. Tick the channels with your instructions, FAQ, terms or product descriptions and click "Save". The project bot becomes a read-only member of a ticked channel, and its published articles become the assistant's knowledge. The bot doesn't post or change anything there. Articles are searched by words and by meaning, and the assistant quotes only the sections it actually read, so clear headings help.

Files and tables. Click "Add file" and upload a file from your device or choose one from Storage.

Kind

Formats

Limits

Documents

PDF, DOCX, TXT, MD, HTML

up to 25 MB per file; a PDF is read up to 500 pages

Tables

XLSX, CSV

up to 20 MB, up to 50,000 rows and 60 columns; the first row is the header

A project can have up to 100 files. A document is split into fragments by its headings. The assistant finds the relevant fragments by words and by meaning, in any language, and remembers where each one came from: the file, the section and the pages or lines. A table is not searched as text: the assistant queries it like a small database. In "Columns" you can rename columns and set their types (Text, Number, Date, Yes/No), and the assistant then filters, sorts and sums by them, for example "the cheapest model in stock". One answer uses at most 20 table rows.

Each file shows its state: "Uploading…", "Waiting for processing", "Processing: N of M fragments", "Ready" or "Error" with the reason. "Replace" uploads a new version of the same file (a document only with a document, a table only with a table), and the assistant keeps answering from the previous version until the new one is ready. "Remove" takes the file out of the knowledge base; the file itself stays in Storage. A scan without a text layer can't be read yet.

Sources under answers. "Show customers the names of source files under answers" adds a line with the file names under each answer. It is off by default. Links to the files are never shown to customers.

Storage space. Knowledge files are kept in Storage and use the project owner's space, like any other file. There is no separate free allowance. A file chosen from your Storage stays where it was and is not copied.

Data connectors

Data connectors let the assistant read live data from your own system, such as an order status, a booking or a balance, through your API. You add a connection once (the base address and the key) and then describe its methods. Connectors only read data, the model never sees your key, and only public HTTPS addresses are allowed. The full guide: Data connectors: the assistant reads live data from your API.

Test chat

Ask a question the way a customer would. The answer uses the same knowledge and connectors but is not sent to anyone. Under the answer, "Sources and tools" shows:

  • the sources used: file, section, pages or table rows;

  • the searches and table queries that ran, and the connector calls with their result;

  • whether the conversation would be handed over to an operator;

  • "The customer sees:" exactly the text the customer would get;

  • the number of model calls and tokens.

Tokens are counted as usual. "Customer details for the test" lets you enter a name, phone and email for connectors that take these values from the request. The test chat needs saved assistant settings, but the assistant itself can stay turned off: prepare the knowledge, check the answers, then turn it on.

Turning it on

Switch on "Virtual assistant" and click "Save settings". From then on the assistant answers new text messages from customers in the project's requests. The Sensever model can be turned on only when it has something to answer from: at least one knowledge base channel, one processed file or one enabled connector method. "Usage over 30 days" under the channels shows input and output tokens by model.

A project without a website

You don't need a website to use the assistant. A seller on a marketplace or a business that only talks to customers on WhatsApp creates a project with the "Project website" field left empty, connects WhatsApp, Telegram, Instagram or MAX in the "Channels" tile and turns the assistant on here. The tile is saved on its own, without any widget settings, and the assistant answers in the messengers. A website can be added to the project later.

Hand-over to your team

The assistant answers only from your materials and doesn't invent prices, features or actions. When the project has its own files or data connectors and the assistant found nothing to answer from, it tells the customer that it is passing the question to an operator, and the request goes to your team as unanswered. A customer can also simply ask for a person. Messages without text, such as a photo on its own, go to your team.

When an agent takes a request (an assignment, auto-assignment, a transfer or their first reply), the assistant is paused in it. The ⏸ / ▶ AI button in the request header, next to the status and "Take chat", switches it back on without leaving the chat: the assistant helps again while the request stays with the agent. Back in the queue, the assistant is on again. Agents can open a conversation and read what the assistant said without taking it over. If the model fails, the customer gets a short notice and the request goes into the normal routing. The routing rules are in Customer support: requests, team and routing.

AI in each channel

Every contact channel of the project (each messenger, the website widget and the contact page) has its own switch "AI answers in this channel". It is on by default. When it is off, requests from that channel go straight to people: neither the Sensever model nor your server's bot answers them. You find the switch in the channel's card (the messenger's settings, the "Widget" tile, the "Contact page" tile) and in the "Channel access" block of the "Team" tile. More: Channel access: who handles requests from each channel.

What the model receives

Answers are written by a language model of xAI (Grok). For each answer the model gets the instructions with the bot's name and personality, the customer's current message with the last four text messages of the request, the fragments of articles and files found for this question, the table rows a query returned, and the response fields of the connector methods it called. The customer's contact details, visit history and other requests are not sent to the model, and neither are your files themselves or your API keys. Text extraction and meaning search for your documents run on Sensever's own servers. Details: AI chat for your website.

Комментарийлер

Авторлор