README

🔗 Demo: yu.net.ua

A floating chat widget on your Magento 2 storefront: the customer searches for products, compares them, and asks questions in plain language, and the AI answers grounded in real catalog data — it doesn't make up an answer on its own. Every conversation is stored and analyzed, so the merchant sees what customers actually ask, which products they can't find, and what they worry about most.

What customers can do

Search products in plain language — "something for my wife as a gift", "running shoes under $80", "a red dress in size 42" — the assistant searches the store's real catalog and answers with specific products, not generic phrases.
Compare and get recommendations — "compare these two models", "what's better for a hiking trip", "show me something similar".
Ask about a specific product — "is it waterproof?", "is it in stock in size 42?" — the assistant knows which page the customer is on and doesn't make them name the product again.
Ask about the store — shipping, payment, returns, warranty, current promotions.
Context awareness — the assistant automatically knows which page the customer is viewing (product, category) without requiring any manual input.

What merchants get — conversation analytics

Admin dashboard: number of conversations, number of messages, average response time, spend by LLM provider and model.
Topic classification — every conversation is automatically tagged by topic (product search, comparison, shipping, payment, returns, etc.), so you can see what's asked about most.
Products customers couldn't find — a list of queries the catalog had no results for: a signal for what to add to your assortment.
Popular questions and products — what customers ask about and which products come up most often.
Business insights — every week the AI analyzes the accumulated statistics on its own and writes up findings in plain language (for example: "customers frequently ask about a product that's out of stock" or "shipping questions increased this week").

How it works

The widget is a lightweight floating chat on the storefront that works on top of any Magento theme (plain HTML/CSS/JS — it doesn't conflict with the theme or add weight to the page). The customer's message goes to the server, and the assistant, when needed, calls into the store catalog through search/comparison/recommendation tools (see Yu_AiSearchEngine) — this keeps the model from "inventing" products or prices that don't exist in the store. Every message — both the customer's question and the assistant's answer — is stored along with tokens, cost, and response time, which feeds the analytics.

The language model is served by a separate module, Yu_AiLlm — you can choose OpenAI, Anthropic Claude, Google Gemini, xAI Grok, Gonka, and others, with automatic failover to a backup provider if one goes down.

Requirements

PHP >= 8.1
Magento 2.4.x with Elasticsearch/OpenSearch configured
The Yu_AiLlm and Yu_AiSearchEngine modules (installed automatically as dependencies)
An API key for at least one supported LLM provider

Installation

composer require yu-dev/module-ai-chat
bin/magento module:enable Yu_AiLlm Yu_AiSearchEngine Yu_AiChat
bin/magento setup:upgrade
bin/magento setup:di:compile
bin/magento cache:flush

The module is installed disabled — the widget won't appear on the storefront until you enable it in the admin panel and configure at least one LLM provider.

Configuration

Stores → Configuration → AI → Chat AI

General — enable the module, widget position on the page.
Chat — widget title, welcome message, example questions, system prompt (tone and style of answers), store facts (things the assistant should always know — e.g. shipping terms), color theme — 10 built-in color themes to choose from, plus the ability to create and save your own theme with custom colors.
Analytics — enable topic classification, automatic business insights, conversation history retention period.
Product Search — a fallback SQL-based search for when the search engine is unavailable.

The LLM provider is configured once, in Stores → Configuration → AI → LLM AI — this section belongs to the Yu_AiLlm module and is shared across all AI modules on the store.

Shared Settings — temperature (0–1: lower is more focused and predictable, higher is more creative; automatically scaled to each provider's own native range) and the maximum number of tokens per response.
Provider: OpenAI / Anthropic Claude / Google Gemini / xAI Grok / DeepSeek — for each provider separately: enabled flag, failover priority (lower value = used earlier in the chain), API key (stored encrypted), model, input/output token price (used only for the cost estimate shown in analytics — update it manually whenever you change the model or rates change).
Provider: Gonka — a decentralized GPU network with open-weight models (Kimi, MiniMax, Qwen); accessed through a broker — same fields as the other providers, plus the broker URL.
Provider: Custom (OpenAI-compatible) — any service with an OpenAI-compatible chat completions API: OpenRouter, Groq, Together, Fireworks, a local Ollama instance, etc. — service URL, key, model.

Providers are tried in priority order: if the active provider fails, the request automatically falls through to the next one.

Author

Yuriy Akishin:

📧 Email: yuriy.akishin@gmail.com
💼 LinkedIn: https://www.linkedin.com/in/yuriyakishin/
💻 GitHub: https://github.com/yuriyakishin

License

MIT

yu-dev / module-ai-chat

Maintainers

Package info

Statistics

Security