The AI Assistant on the AXL platform is a full-fledged virtual employee powered by advanced large language models (LLMs) that works inside your account 24/7. It automatically communicates with users and clients, answers questions based on your materials, qualifies leads, processes requests, and guides the dialogue to the desired outcome without human intervention.
The assistant works just like a live operator: it perfectly understands the context, maintains a coherent conversation, asks clarifying questions, and offers ready-made solutions, but does so instantly and around the clock.
All management, chat history, and analytics are located within a single AXL ecosystem. You can create 20 or more independent assistants for different tasks.
Click on the desired section of the article to go to its description:
How to connect the AI Assistant
Go to the "Applications" section. In the "All Applications" tab, a list of all available applications for connection to your account will be displayed.
In the "AI Assistant" application card, click the "Connect" button.
In the window that opens, fill in the fields:
Title. Enter a name for the application to make it easier to find; this title is for your eyes only.
Enabled. If this toggle is turned off, the AI Assistant will not reply to messages in connected chat channels.
Display name. Enter the name that will be displayed to users in chats (for example: "Anna from Support", "Course Bot Assistant").
Instructions. Enter instructions regarding the tone, role, and additional rules.
Strategy. Choose how the AI Assistant will balance speed, cost, and analysis quality.
Presets. Add predefined prompt extensions that modify the assistant's behavior in specific situations.
Knowledge base. Add text or attach files (txt, pdf, docx, json) containing the knowledge that the AI Assistant will use when responding.
Analyze attached files. Analyze files attached by users (images and documents). Enabling this option may increase usage costs.
Chat channels. Select the chat channels where the AI Assistant will respond to incoming messages.
Click the "Connect" button.
Workflow logic
On the client (guest) side:
Familiar interface: The client writes to the platform's internal chat or a messenger just like before.
Presence effect: While the response is being generated, a "typing..." status is displayed, creating the feeling of communicating with a real person.
Speed and results: The client receives a detailed and accurate answer within seconds.
Smart escalation of complex questions to a staff member: If a situation goes beyond the loaded instructions, the bot will automatically transfer the dialogue to a live person.
On the administrator (account) side:
One-time setup: You set the name, prompt (instructions), and upload the knowledge base once.
Flexible distribution: You choose which connected communication channel (Telegram, internal chats, etc.) to assign a specific assistant to. You can link an assistant to any connected chat channel in the system. You can also configure unique standalone assistants for individual channels.
All-in-one: Anti-spam system, automatic dialogue transfer, message history, and analytics.
Testing
Go to the "Applications" section. The "Connected" tab will display the "AI Assistant" you have already added. Click the "Manage" button and go to the "AI Assistant" application.
Testing the AI Assistant's responses is available within the "AI Assistant" application.
Testing is a built-in dialogue simulator that allows you to fully test the AI Assistant in a safe environment before giving it access to chats with real users and clients. In testing, the actual cost of the AI Assistant's response in tokens is instantly visible.
The tool works in real time and helps perfect the bot's behavior without the risk of sending an incorrect response to a client. If the AI replied in the wrong tone or missed an important detail, you can change the instruction right here, turn behavior presets on or off to alter its behavior in specific situations, and immediately send a repeat request.
Instructions for the AI Assistant
You can set custom instructions for each assistant, defining the style and rules of the assistant's messages.
Examples of instructions:
communication style with the client
message tone (strict, neutral, friendly)
permissibility of jokes or informal explanations
other instructions
The AI Assistant strictly follows the given instructions when responding to clients.
Ready-made behavior presets
You can combine ready-made assistant behavior templates using simple switches (checkboxes):
Avoid sensitive topics: It will gracefully steer away from discussions about politics, religion, territorial disputes, and other sensitive topics.
Format the answers nicely: Formats responses using structured layouts - lists, paragraphs, and spacing - for better readability.
Step-by-step information entry: Helps conduct dialogue in interview mode — the assistant asks exactly one question at a time, without overwhelming the client with all items at once.
Knowledge base
In the AI Assistant settings, you can upload your own knowledge base to be used when responding to clients. This can include: your course, webinar, and intensive programs; frequently asked questions (FAQs) and internal account regulations; price lists, instructions, tariff grids, and installment terms.
Various material formats are supported:
text files
PDF
- DOCX
- JSON
other document formats
Using a knowledge base allows you to respond to clients taking into account your terminology and internal standards, rather than using template formulations.
Attachment analysis
The AI Assistant perfectly understands formats other than plain text:
Images: analyzes images and understands what is on them.
Documents: reads, parses, and takes into account files attached and sent by the user during the dialogue.
Voice messages: transcribes audio into text, analyzes the meaning, and replies in the chat.
Handover to a live employee
The assistant knows when to step back and understands when it is better to transfer the dialogue to a live employee. This happens if:
The question turns out to be too complex, non-standard, or goes beyond the scope of the knowledge base.
The user directly and explicitly asks to call a human (the "Call a human" exit path is built into the system by default).
Any other special condition specified by you in the instructions is triggered.
Built-in anti-spam
Before sending a request to the main (more expensive) language model, the message is analyzed by a separate "lightweight" and inexpensive LLM. If the system detects obvious spam, the assistant recognizes the attack, automatically pauses responses for this user for 15 minutes, and saves your tokens.
Automation via scenarios
The AI Assistant is integrated into all platform scenarios and functions there as an individual step:
You can add an "AI Assistant Chat" node to any of your scenarios.
Communication with the user inside the node runs cyclically until the goal is achieved.
You can configure dialogue "exits" based on user intents. The assistant itself will understand which outcome the conversation has reached (for example: "Application sent", "Issue resolved", "Refusal", "Ready for the next step"), end the cycle, and direct the contact along the corresponding branch of the scenario.
In scenarios, you can build sequential chains consisting of several different assistants.
"AI Assistant Chat" Node Settings:
- Node name. Specify the name of the node; you can name the nodes in the scenario at your own discretion.
- AI Assistant. Select one AI Assistant from the AI helpers previously created in the system.
- Additional instructions. A field for entering contextual guidelines that complement or temporarily adjust the assistant's main system prompt (instruction) specifically for this step of the scenario.
- Time to end chat. A parameter determining the user inactivity timeout. If the client stops responding, then after the specified time has elapsed (for example, 5 minutes), the chat will automatically close, and the contact will exit the scenario node via the "Chat Completion" technical exit.
- Outputs. The "smart" routing configuration block. Here you create custom outputs from the node based on client intents. For each output, a name and a condition are specified. Name is the name of the output button on the scenario diagram. Condition is a natural language logical rule by which the AI Assistant will understand that the goal has been achieved.
- Testing. A built-in dialogue simulator that allows you to fully test the AI Assistant in a safe environment before giving it access to chats with real users and clients. In testing, the actual cost of the AI Assistant's response in tokens is instantly visible.
Strategy selection and cost
You can flexibly manage your budget by choosing a working strategy for each specific assistant depending on its tasks:
Faster and cheaper - lightweight model, instant responses, minimal cost.
Optimal - a balance of quality and cost, suitable for most tasks.
- More effective and more expensive - a top model for complex dialogues and sales.
When changing the strategy, the price per dialogue changes significantly — you choose according to your use case.
Cost
Fixed part: $5/day subscription fee for the module (allows you to create and enable an unlimited number of assistants without restrictions).
Variable part: Pay-as-you-go for used tokens.
