gemini_file_search

Author: r3-yamauchi
Version: 0.0.1
Type: tool

Description

brings Google's Gemini File Search capabilities to your Dify workflows. This plugin provides nine focused tools for managing file search stores and performing semantic search with RAG (Retrieval-Augmented Generation):

Store Management: Create, list, view, and delete file search stores
Document Ingestion: Upload files with automatic chunking and embedding
Document Visibility & Hygiene: List, inspect, and remove the files that live inside a store
Semantic Search: Query indexed documents with AI-generated answers and citations

The source code of this plugin is available in the GitHub repository.

Features

Tier 1: Core RAG Functionality

Create Store (): Initialize new file search stores with optional display names
Upload File (): Upload and import files directly with automatic chunking, embedding, and indexing
Search (): Perform semantic search across indexed documents with AI-generated answers and citations

Tier 2: Store Management

List Stores (): View all file search stores in your project
Get Store Details (): Inspect specific store properties and metadata
Delete Store (): Remove stores and their data permanently

Tier 3: Document Visibility

List Files (): Enumerate the documents stored in a given file search store with paging controls
Get File Details (): Retrieve metadata, ingestion state, and timestamps for a single document
Delete File (): Remove individual documents without touching the rest of the store

Key Capabilities

Automatic Chunking: Files are automatically split into optimal chunks for embedding
Custom Metadata: Tag documents with searchable metadata (author, date, category, etc.)
Metadata Filtering: Search within document subsets using filter expressions
Citation Support: AI answers include references to source documents
Configurable Chunking: Customize token limits and overlap for fine-grained control
Multi-Model Support: Works with gemini-2.5-flash and gemini-2.5-pro
Synchronous Operation: All long-running operations poll automatically until completion

Prerequisites

Dify SaaS or self-hosted instance with plugin support enabled
Google Gemini API key from Google AI Studio
Supported file formats: PDF, TXT, DOCX, Markdown, code files, and many more

Installation & Setup

Install the plugin from the Dify Marketplace or upload this package manually
Configure provider credentials:
- Gemini API Key (required): Your API key from Google AI Studio
- Default Model (optional): Defaults to
Add tool nodes to your workflow and configure parameters as described below

Tool Parameters

1. Create File Search Store ()

Creates a new file search store for storing document embeddings.

Parameter	Required	Description
	❌	Optional human-readable name for the store

Returns: Store name (e.g., ), display name, creation timestamp

2. Upload File ()

Upload and import a file with automatic chunking and indexing.

Parameter	Required	Description
	✅	Target store name (e.g., )
	✅	File to upload (PDF, TXT, DOCX, Markdown, etc.)
	✅	Unique name for citations (visible in search results)
	❌	Metadata as JSON array or direct list/dict (e.g., )
	❌	Maximum tokens per chunk (custom chunking)
	❌	Overlapping tokens between chunks (custom chunking)

Returns: Operation status, store name, display name, completion details, and (document resource name) for downstream tools

Example metadata (can be provided as JSON text or list objects):

Note: Gemini File Search currently enforces a 100 MB per-file limit. The upload tool validates this locally and will raise an error before calling the API if a file exceeds the limit.

3. Search ()

Perform semantic search with AI-generated answers.

Parameter	Required	Description
	✅	Store to search (e.g., )
	✅	Search query or question
	❌	Model override (defaults to provider setting)
	❌	Filter expression (e.g., )
	❌	Include citation information (default: )

Returns: AI-generated answer, citations with source references, grounding metadata

Example filter syntax:

- Exact match
- Numeric match
- Combined filters

4. List Stores ()

List all file search stores in your project.

No parameters required

Returns: Array of stores with names, display names, and creation timestamps

5. Get Store Details ()

Get detailed information about a specific store.

Parameter	Required	Description
	✅	Store name to retrieve

Returns: Store name, display name, creation time, update time

6. Delete Store ()

Delete a file search store and all its data permanently.

Parameter	Required	Description
	✅	Store name to delete
	❌	Force deletion even if store contains documents (default: )

Returns: Deletion status

7. List Files ()

List the documents that currently reside in a specific file search store.

Parameter	Required	Description
	✅	Store that owns the documents (e.g., )
	❌	Optional page size between 1 and 100. Defaults to the API setting.
	❌	Pass the from a previous response to continue listing.

Returns: Array of document metadata (name, display name, state, size, timestamps), count, and for pagination

8. Get File Details ()

Retrieve metadata and ingestion status for a single document.

Parameter	Required	Description
	✅	Store containing the document
	❌	Full document resource name as returned by
	❌	Document identifier when you do not have the full resource name (with or without the prefix)

Note: Provide either or . If both are supplied, takes precedence.

Returns: Document metadata including ingestion state, file size, timestamps, and custom metadata

9. Delete File ()

Delete one document inside a store without removing the entire store.

Parameter	Required	Description
	✅	Store containing the document
	❌	Full document resource name (preferred). Overrides when supplied.
	❌	Document identifier when you do not have the full resource name
	✅	Must be set to to prevent accidental deletions
	❌	Force deletion even if the document is in use

Returns: Deletion status with store name, document name, and whether force mode was used

Workflow Examples

Example 1: Document Upload and Search

Example 2: Filtered Search with Metadata

Example 3: Store Management

Supported File Formats

Gemini File Search supports a wide range of file formats including:

Documents: PDF, DOCX, ODT, TXT, Markdown, RTF
Spreadsheets: XLSX, CSV, TSV
Presentations: PPTX
Code: Python, JavaScript, Java, C++, Go, Rust, and many more
Data: JSON, XML, YAML, SQL

See the Gemini API documentation for the complete list.

Rate Limits & Quotas

Max file size: 100 MB per document
Stores per project: 10
Total storage (tier-dependent):
- Free: 1 GB
- Tier 1: 10 GB
- Tier 2: 100 GB
- Tier 3: 1 TB
Recommended store size: < 20 GB per store for optimal retrieval latency

Pricing

Indexing: $0.15 per million tokens (based on embedding pricing)
Storage: Free
Query-time embeddings: Free
Retrieved document tokens: Charged as normal context tokens

Troubleshooting

Error	Solution
API key invalid	Verify your Gemini API key in provider settings
Quota exceeded	Check your API usage limits and upgrade tier if needed
Store not found	Confirm the store name exists using
File upload timeout	Large files may take time; operation polling handles this automatically
Metadata filter syntax error	Use correct syntax: for strings, for numbers

Privacy Policy

See PRIVACY.md for detailed information about data handling and security.

Data Handling

This plugin processes:

Files: Uploaded documents are sent to Gemini API for chunking and embedding
API Credentials: Gemini API key is stored securely in plugin settings
Store IDs: File search store names are used to organize embeddings

Storage

Uploaded files (via Files API) are deleted after 48 hours
File search store data persists until manually deleted
No data is stored locally by the plugin

Security

All API communications use HTTPS encryption
API keys are stored securely by Dify's credential management system

License

This project is released under the MIT License.