What this solution does
The service solves business tasks using generative models. All requests are logged, so entering personal or confidential data is prohibited. Publishing model responses with mentions of Indexis is forbidden. Using the service to violate laws or mislead users is not allowed. The generator may provide incorrect information, and its responses do not reflect the developer's position. Review the usage rules before starting. Once created, the experiment becomes available to all catalog participants. Model requests are billed.
The module adds an AI chat with Retrieval-Augmented Generation (RAG) support to 1C-Bitrix websites. Visitors ask questions, and the system answers by analyzing uploaded files (PDF, DOCX, XLSX) as well as data from information blocks, highload blocks, and trade offers.
Installation creates the components indexis:ai.chat.documents (providing a full-screen chat with history or a modal window widget) and indexis:callback.form (a lead capture form). The module also generates highload blocks for storing chats and messages and adds a configuration page to the administrative section.
AI chat with conversation history: chats and messages are stored in highload blocks, and the visitor history retention period is configurable.
RAG based on your documents: PDF, DOCX, and XLSX files are uploaded, converted to text, and indexed.
Indexing of information blocks, highload blocks, and trade offers into separate Qdrant collections.
Qdrant database management from the admin panel: displays a list of indexed documents with the indexing date, supporting single or mass deletion.
Indexing configuration: allows selection of the embedding model, vector size, chunk size, and chunk overlap.
Integration with an external AI provider via API key (KEY_OPEN_ROUTER): enables selection of primary and backup models, plus configuration of the general prompt and assistant response style.
Target actions: the assistant suggests specific steps to the visitor (such as submitting a request) according to your instructions.
Conversation transfer to a manager in Telegram: the manager replies within the group thread, and the response appears in the site chat.
Voice input: speech recognition is performed via Yandex SpeechKit.
Bot protection: implemented using Yandex SmartCaptcha.
Personal data masking: phone numbers, email addresses, passport data, and full names are masked before being sent to external AI services.
The module limits the number of daily requests to the neural network using visitor data via the AiVisitedUsers highload block.
It allows customization of the chat interface, including button, message, and menu colors, as well as the chat logo and widget icon.
It supports operation through a proxy server for AI requests and for Telegram integration.
If the connection to the AI provider or Telegram is unavailable, configure the proxy in the module parameters.
The accuracy and quality of the system's answers depend directly on the completeness of the information in the uploaded files.
Operation requires a 1C-Bitrix edition that supports highload blocks.
The PHP cURL extension and an SSL certificate on the site for HTTPS protocol operation are required.
A server-accessible Qdrant service or a compatible service with a valid API key is required.
An OpenRouter API key or a key from another provider compatible with the used API is required.
For additional functions, the following are needed: a Telegram bot token and a group with topics, a Yandex Cloud API key and catalog ID for speech recognition, and Yandex SmartCaptcha keys.
General iT can connect “AI consultant for Bitrix. Chat based on your documents and neural networks.”, install and configure the module, and make sure the integration works correctly.
