n8n-nodes-file-search

This is an n8n community node for Google File Search (Gemini API RAG).

Google File Search is a fully managed RAG (Retrieval Augmented Generation) system that allows you to store, index, and semantically search documents using the Gemini API.

n8n is a fair-code licensed workflow automation platform.

Installation

Follow the installation guide in the n8n community nodes documentation.

Operations

Store

Manage File Search stores:

• Create - Create a new File Search store
• List - List all File Search stores
• Get - Get details of a specific store
• Delete - Delete a store (with force option for non-empty stores)

Document

Manage documents within stores:

• Upload - Upload a file directly to a store (supports binary data from previous nodes)
• Import - Import an existing file from the Files API into a store
• List - List all documents in a store
• Get - Get document details and status
• Delete - Delete a document from a store

Features:

• Wait for Completion: Polls until the document is fully indexed
• Metadata: Attach custom key-value pairs for filtering during queries
• Chunking Options: Configure max tokens per chunk and overlap

Query

Query stores with semantic search:

• Search - Query one or more stores and generate a response using a Gemini model
• Deep Research - Run the Deep Research agent for comprehensive, multi-step analysis

Search Features

• Multiple Stores: Query across multiple stores at once
• Metadata Filter: Filter results using AIP-160-like syntax (e.g., category = "reports" AND status = "published")
• Structured Output: Get responses in a specific JSON format using a JSON Schema (Gemini 3+ models)
• System Prompt: Provide custom system instructions
• Model Selection: Choose between Gemini 3 Flash/Pro, 2.5 Flash/Pro, or 2.0 Flash
• Grounding Metadata: Include source citations in the response

Deep Research Features

• Autonomous Research: The agent plans, searches, reads, and iterates to produce detailed reports
• Web + File Search: Combines public web search with your File Search stores (optional)
• Output Formatting: Steer the output format with custom instructions
• Long-running Tasks: Automatically polls for completion (typically 5-20 minutes)

Credentials

This node requires a Google Gemini API credential with an API key from Google AI Studio.

Usage Examples

Upload and Query Workflow

1. Convert to File node - Convert JSON/data to a binary file
2. Google File Search (Document → Upload) - Upload to your store
3. Google File Search (Query → Generate Content) - Query the store

Batch Processing

The node supports processing multiple items. Use with:

• Loop nodes for batch uploads
• Conditional nodes to handle different document types
• Set node to prepare metadata before upload

Metadata Filtering

When querying, you can filter results using metadata:

category = "reports"
status = "published"
year >= 2024 AND department = "engineering"

Note: Metadata must be attached during document upload/import to be filterable.

Structured Output

Starting with Gemini 3 models, you can get responses in a specific JSON format by enabling Structured Output and providing a JSON Schema:

{
	"type": "object",
	"properties": {
		"summary": { "type": "string", "description": "A brief summary of the findings" },
		"key_points": { "type": "array", "items": { "type": "string" } },
		"confidence": { "type": "number", "description": "Confidence score 0-1" }
	},
	"required": ["summary", "key_points"]
}

This is useful for:

• Data extraction: Pull specific information from documents
• Structured classification: Categorize content with structured labels
• Integration workflows: Get predictable, parseable outputs for downstream nodes

See the Gemini Structured Output documentation for supported schema properties.

Deep Research

The Deep Research operation uses the Gemini Deep Research Agent to autonomously research complex topics. Unlike the quick Search operation, Deep Research:

• Takes minutes (not seconds) to complete
• Produces detailed reports with citations
• Can combine web search + your documents
• Costs approximately $2-5 per task

Example Use Cases

• Market analysis and competitive landscaping
• Due diligence and literature reviews
• Comparing internal documents against public information
• Technical research and trend analysis

Usage

1. Select Query → Deep Research
2. Enter your research query (be specific about what you want to learn)
3. Optionally enable Include File Search Stores to add your documents as sources
4. Optionally add Output Format Instructions to structure the report

Research the competitive landscape of EV batteries.

Format the output as a technical report with:
1. Executive Summary
2. Key Players (include a comparison table)
3. Technology Trends
4. Supply Chain Risks

Follow-up Questions

After research completes, you can ask follow-up questions without restarting the entire research:

1. Store the interactionId from the first Deep Research output
2. Run Deep Research again with a follow-up question
3. Provide the previous interactionId in the Previous Interaction ID field

This lets you ask for clarification, summarization, or elaboration on the report.

Note: Deep Research is in preview and may take 5-20 minutes to complete. The node will poll automatically until the research is done or the max wait time is reached.

Resources

• Google File Search Documentation
• Gemini Deep Research Documentation
• Gemini API Overview
• AIP-160 Filtering
• n8n Community Nodes

License

MIT