---
title: "Models & Prompts"
id: "1836"
type: "page"
slug: "models"
published_at: "2026-09-20T23:49:51+00:00"
modified_at: "2026-10-22T16:38:23+00:00"
url: "https://xedant.com/agents/chatbot/docs/models/"
markdown_url: "https://xedant.com/agents/chatbot/docs/models.md"
excerpt: "AI in Chatbot Agent means calls to an external model. This is where you configure…"
---

# Models & Prompts

[https://xedant.com/agents/chatbot/docs/models/.md](https://xedant.com/agents/chatbot/docs/models/.md)

AI in Chatbot Agent means calls to an external model. This is where you configure which service to call, which model answers by default, which ready-made texts are used, and how much it all costs.

## Providers

You can connect any service compatible with OpenAI — that is, most modern AI services. You set the service address, the list of models with prices, and the access keys. The keys are stored on the server and always shown masked in the interface: just a hint that a key is set. If you save an already masked value, the key does not change — so it cannot be erased by accident.

A special case is a provider made with a Python script. It saves the day when the model lives inside your network and cannot speak the common language of AI services: the script receives the whole request and decides itself how to fulfill it. Such a script goes into the folder shared by the whole application, not with an individual bot; details in the [Scripts](/agents/chatbot/docs/scripts)
 section.

## Choosing a model

The model is chosen along a chain: first the specific scenario node is checked; if nothing is set there — the bot’s setting; if that is missing too — the platform-wide choice. So one bot can answer with a cheap model and use a stronger one for the hard cases. Fallback models can be set too: if the first one fails to answer, the next one replies, and the step’s trace records who exactly answered.

## Prompts in one place

Prompts are ready-made instruction texts for AI. They are stored in one place, and nodes of different bots can refer to them. Edit the text once — the behavior of every bot that references it changes. A prompt’s name cannot be renamed: to rename, delete the old one and create a new one. That way references never drift.

## The connection test

Every model has a test button: the application makes a real request and shows the result and the response time. Useful right after setup — to make sure the key is correct and the service responds.

## Spend tracking

Every AI call is recorded: which bot, which model, how many tokens went in and out, what it cost, how long it took and whether it succeeded. A token is the conventional unit of text length that AI services bill by. The report shows spending for a period, broken down by bot and model — so it is clear where the money goes.

## Rules for a bot

Constraints are a set of instructions about tone, prohibitions and the boundaries of the conversation. They are added to every AI call the bot makes, so a uniform style of communication is kept across all nodes at once. For example: “Never invent prices — say only what is in the price list” or “Do not promise deadlines that are not in the documents”.

Next: how to answer from your documents — in the [Knowledge Base](/agents/chatbot/docs/knowledge)
 section, how to use AI in the scenario — in [Flow Nodes](/agents/chatbot/docs/nodes)
, and the numbers on spending — in [Analytics & Logs](/agents/chatbot/docs/analytics)
.

[← Back to the documentation index](/agents/chatbot/docs)