Implement RAG Tools for AI Agents

Problem/Motivation

We need to integrate Typesense with the AI module's "agent" functionality to enable Retrieval Augmented Generation (RAG) in the chatbot. The existing ai_search module provides a RagTool that can trigger RAG, but it only supports specific vector databases like Milvus, Zilliz, and Pinecone. We need to extend this functionality to work with Typesense, leveraging its vector database capabilities.

Proposed resolution

Historical proposal (kept as history): The proposed resolution was to add a FunctionCall plugin to the existing Typesense module. This plugin would enable an agent to perform queries on Typesense, allowing us to leverage its vector database capabilities for RAG. This approach made sense previously because it kept Typesense-specific logic within its own module and avoided changing the core AI search module.

Updated proposal (use Tool API): Rather than implementing a FunctionCall plugin, implement a Tool plugin for the Typesense module using the Tool API module (https://www.drupal.org/project/tool). The Typesense Tool should expose a concise, well-documented tool contract that agents can call to perform semantic/vector searches and return results in a small, predictable JSON structure suitable for RAG.

Remaining tasks

• [x] Analyze the ai_search module to understand how FunctionCall plugins are implemented and integrated with agents.
• [x] Develop the FunctionCall plugin within the Typesense module to handle vector search queries.
• [x] Ensure the plugin successfully retrieves relevant content from Typesense and provides it to the AI agent for RAG.
• [x] Replace the FunctionCall plugin with Tool

User interface changes

None. The changes are all backend-related.

API changes

None.

Data model changes

None. The changes only affect how data is retrieved, not how it is stored.