Nevred RAG Chat for WooCommerce

Сүрөттөө

Nevred RAG Chat for WooCommerce adds an AI-powered chat widget to your site that answers visitor questions based on your own content: WooCommerce products, posts, and pages. It implements a full Retrieval-Augmented Generation (RAG) pipeline with a choice of two AI engines:

• Ollama (self-hosted, free) — everything runs on your own infrastructure; no data leaves your servers and no paid APIs are required.
• OpenAI API — enter your API key and the plugin uses OpenAI models for embeddings and chat answers; no local AI server needed.

The vector database can be a self-hosted Qdrant instance or a managed Qdrant Cloud cluster (a free tier is available) — just enter its URL and API key.

How it works:

1. Your content (products, posts, pages) is converted into vector embeddings by the selected engine (e.g. nomic-embed-text or bge-m3 in Ollama, or text-embedding-3-small via OpenAI).
2. The vectors are stored in a Qdrant vector database (self-hosted or Qdrant Cloud).
3. When a visitor asks a question, the plugin finds the most relevant content by semantic similarity and passes it as context to the chat model (e.g. qwen2.5:1.5b in Ollama, or gpt-4o-mini via OpenAI).
4. The model generates an answer with real links to your products and pages.

Features:

• Two AI engines: self-hosted Ollama or the OpenAI API (just enter your API key).
• Qdrant Cloud support: connect a managed cluster with an API key — no servers to maintain.
• Floating chat widget with a customizable primary color.
• Semantic (vector) search over products, posts, and pages.
• WooCommerce support: product SKU, price, categories, and attributes are indexed.
• Automatic re-indexing when content is created, updated, or deleted (via WP-Cron).
• Bulk indexing from the admin page or via WP-CLI (wp nevredrag index).
• Configurable system prompt, context template, and product card template.
• Optional “no-AI” mode: the bot replies with product cards only, without LLM generation.
• WP-CLI commands for testing: wp nevredrag search "query" and wp nevredrag ask "question".
• Translation-ready (Ukrainian and Russian translations included).

External services

Depending on your configuration, this plugin connects to the following services. With the default configuration (Ollama + self-hosted Qdrant) no data is sent to third-party companies.

Ollama (local LLM server) — used when the “Ollama” engine is selected (default).

• What it is used for: generating text embeddings for your content and generating chat answers.
• What data is sent: the text of your published products, posts, and pages (during indexing), and visitor chat messages together with the retrieved site context (during chat).
• When: during content indexing and on every chat message.
• Service link: https://ollama.com — runs on your own server; no external terms of service apply.

OpenAI API (third-party service) — used only when you select the “OpenAI API” engine and enter your API key.

• What it is used for: generating text embeddings for your content and generating chat answers.
• What data is sent to OpenAI: the text of your published products, posts, and pages (during indexing), and visitor chat messages together with the retrieved site context (during chat), along with your API key for authentication.
• When: during content indexing and on every chat message.
• Service: https://openai.com — Terms of use, Privacy policy. Usage is billed to your OpenAI account.

Qdrant (vector database, self-hosted or Qdrant Cloud)

• What it is used for: storing and searching vector embeddings of your content.
• What data is sent: vector embeddings and text excerpts of your published content, query vectors built from visitor chat messages, and your API key (if configured) for authentication.
• When: during content indexing and on every chat message.
• Service: https://qdrant.tech. Self-hosted instances run on your own server. If you connect a managed Qdrant Cloud cluster, the data above is stored on Qdrant’s infrastructure — see the Qdrant Cloud terms and privacy policy.