AI RAG Search Chat

AI RAG Search Chat adds an AI-powered search page and a full conversational chat interface to your Drupal site using Retrieval-Augmented Generation (RAG). Users ask questions in natural language and receive answers grounded in your site's own content, with linked sources.

Features

• AI Search page (/ai-search): A single-query interface where users type a question and receive an AI-generated answer with a list of source content from your site. Works like a smarter search box that understands meaning, not just keywords.
• AI Chat page (/ai-search/chat): A full conversational chat interface with persistent session history. Users can have multi-turn conversations, switch between sessions, and pick up where they left off.
• Retrieval-Augmented Generation: Queries your Search API vector index for semantically relevant content chunks. These are sent as context to the LLM, so answers are grounded in your actual content rather than hallucinated.
• Session management: Chat sessions are stored in the database with full message history. Sessions can be searched, listed, and deleted. Session titles are auto-generated after the first message.
• Anonymous user support: Anonymous visitors get their own isolated sessions via a secure cookie, with a separate, shorter retention period.
• Rate limiting: Configurable per-role rate limits for messages and session creation, powered by Drupal's Flood API. Works with any flood backend (database, Redis, etc.).
• Server-rendered components: All UI is built with Drupal Single Directory Components (SDC). API endpoints return pre-rendered HTML, so the frontend inserts markup without client-side templating.
• Page cache friendly: Static page shells are cached by Varnish or any reverse proxy. Personalized content loads via AJAX after page render.
• Provider agnostic: Uses the Drupal AI module's provider system. Works with any supported LLM (OpenAI, Anthropic, etc.) and any supported vector database (Milvus, Pinecone, Qdrant, etc.).
• Configurable system prompt: Customize the instructions sent to the LLM to control tone, format, and behavior.
• Source deduplication and pagination: Sources are deduplicated server-side and paginated with a "Load more" button.
• Async title generation: Session titles are generated via a queue worker using a separate (optionally lighter/cheaper) model so they do not block the chat response.

Post-Installation

After enabling the module:

1. Set up your AI infrastructure: You need a configured LLM provider and a vector database. Navigate to Administration > Configuration > AI > AI Settings to configure providers.
2. Create a Search API server and index: Add a server using the AI Search backend, then create an index for the content types you want searchable. Index your content with drush search-api:index.
3. Configure the module: Go to Administration > Configuration > AI > AI RAG Search Chat (/admin/config/ai/ai-rag-search-chat). Select your vector search index(es), choose LLM providers for chat and title generation, customize the system prompt, and adjust rate limiting, history, and display settings.
4. Set permissions: Grant "Access AI RAG Search Chat" to the roles that should use the feature. Grant to Anonymous if you want unauthenticated access.
5. Visit /ai-search to test a query and /ai-search/chat to verify the chat interface.

Additional Requirements

• AI (Artificial Intelligence) - Core AI provider abstraction
• AI Search - Search API backend for vector search
• Search API - Search framework
• Key - Secure API key storage
• A VDB (vector database) provider module for your chosen database (e.g., Milvus, Pinecone, Qdrant)
• An LLM provider API key (e.g., OpenAI, Anthropic)
• An embedding provider API key (may be the same as the LLM provider)

Supporting this Module

This module was developed and is maintained by Kanopi Studios.

Community Documentation

See the module's included README.md for detailed architecture documentation, API endpoint reference, database schema, and configuration instructions.