This plugin provides Google Gemini integration for Fess's RAG (Retrieval-Augmented Generation) features. It enables Fess to use Google Gemini models for AI-powered search capabilities including intent detection, answer generation, document summarization, and FAQ handling.

See Maven Repository.

• Fess 15.x or later
• Java 21 or later
• Google Gemini API key

1. Download the plugin JAR from the Maven Repository
2. Place it in your Fess plugin directory
3. Restart Fess

For detailed instructions, see the Plugin Administration Guide.

Configure the following properties in fess_config.properties:

The plugin authenticates by sending the API key as the x-goog-api-key HTTP request header (Google's recommended method). Keys are never appended to the URL as ?key=…, so they do not appear in URL access logs.

The plugin automatically translates a single request-level thinkingBudget (integer token allowance) to whatever shape the resolved model expects:

• Gemini 2.x (e.g. gemini-2.5-flash) – sent as thinkingConfig.thinkingBudget (integer).
• Gemini 3.x (e.g. gemini-3-flash-preview, gemini-3.1-flash-lite-preview, gemini-3.1-pro) – sent as thinkingConfig.thinkingLevel with bucket mapping <=0 → MINIMAL (Flash / Flash-Lite) or LOW (Pro, which does not support MINIMAL), <=4096 → MEDIUM, >4096 → HIGH.

Thinking parts (response parts marked thought: true) are automatically filtered out before the visible response is delivered. Override per prompt type via rag.llm.gemini.<type>.thinking.budget (and rag.llm.gemini.<type>.max.tokens for the visible-output cap).

• Intent Detection - Determines user intent (search, summary, FAQ, unclear) and generates Lucene queries
• Answer Generation - Generates answers based on search results with citation support
• Document Summarization - Summarizes specific documents
• FAQ Handling - Provides direct, concise answers to FAQ-type questions
• Relevance Evaluation - Identifies the most relevant documents for answer generation
• Streaming Support - Real-time response streaming via Server-Sent Events (?alt=sse)
• Availability Checking - Validates API availability at configurable intervals
• Extended Thinking - Model-aware thinking config: integer thinkingBudget for Gemini 2.x and bucketed thinkingLevel (MINIMAL/LOW/MEDIUM/HIGH) for Gemini 3.x

The API key is supplied via the x-goog-api-key request header on every call (it is not appended to the URL).

• GET /models - Lists available models for availability checking
• POST /models/{model}:generateContent - Performs chat completion
• POST /models/{model}:streamGenerateContent?alt=sse - Performs streaming chat completion (Server-Sent Events)

mvn clean package

mvn test

Apache License 2.0