No. This is the most important design decision in the architecture. The bot only answers product questions using data retrieved from the Qdrant vector database — which is a direct sync of your Bitrix24 product catalogue. GPT-4o-mini is instructed to use only the retrieved context and to decline to answer if no matching product is found. It does not draw on general training knowledge for product facts. If the product isn't in your catalogue, the bot says so.

The bot uses the Bitrix24 imbot API and Open Channels — both available on Bitrix24 Self-Hosted (On-Premise) and Cloud plans that include the CRM and Open Channels features. Standard and Professional Cloud plans include Open Channels. Self-Hosted Enterprise plans have full access. We confirm compatibility with your specific plan before starting implementation.

A full replace sync runs automatically every 24 hours via n8n, pulling the latest product catalogue from Bitrix24 and re-embedding everything into Qdrant. Mid-day price or stock changes won't be reflected until the next sync cycle. For businesses where real-time accuracy is critical, we can configure more frequent syncs (e.g. every hour) — the cost scales with OpenAI embedding usage, which remains very low (roughly $0.002 per 2,000-item sync).

Yes. The OpenAI text-embedding-3-small model supports multilingual queries natively — a customer typing in Arabic, French, or Spanish will still match against English product names in the vector store. GPT-4o-mini can respond in the customer's detected language. For Standard and Enterprise packages, we configure language detection and dual-language product descriptions if your catalogue is multilingual.

The n8n workflow has a Customer Config node at the top — a single JSON object containing the customer's Bitrix24 credentials, their Qdrant collection name, their OpenAI API key, the bot personality setting, and the escalation rules. Deploying the bot for a new customer means cloning the workflow, updating this one node, and running the initial sync. Each customer's vector data is stored in a fully isolated Qdrant collection — there is no sharing of product data between deployments. This makes it ideal for agencies running the bot as a service for multiple clients.

The bot is configured with escalation triggers — when it detects a request outside its scope (approval for a custom price, a complaint, a question it can't answer from the catalogue), it sends a notification to the assigned Bitrix24 agent in the same chat thread. The customer sees a message like "I'll get a team member to assist you with that" and the human takes over seamlessly. The entire conversation history is visible to the agent in Bitrix24, so there's no need to repeat context.