feat: Gemini file search vectordb

cookbook/knowledge/vector_db/README.md

@@ -34,6 +34,7 @@ agent = Agent(
 - **[ChromaDB](./chroma_db/)** - Embedded vector database
 - **[ClickHouse](./clickhouse_db/)** - Columnar database with vector functions
 - **[Couchbase](./couchbase_db/)** - NoSQL database with vector search
+- **[Gemini File Search](./gemini_file_search/)** - Google's managed vector database with Gemini integration
 - **[LanceDB](./lance_db/)** - Fast columnar vector database
 - **[LangChain](./langchain/)** - Use any LangChain vector store
 - **[LightRAG](./lightrag/)** - Graph-based RAG system

cookbook/knowledge/vector_db/gemini_file_search/README.md

@@ -0,0 +1,109 @@
+# Gemini File Search
+
+Gemini File Search provides a managed vector database integrated with Google's Gemini AI models. It allows you to upload documents and perform semantic search using Google's infrastructure.
+
+## Features
+
+- **Managed Service**: No infrastructure management required
+- **Native Integration**: Seamlessly works with Gemini models
+- **File Upload**: Directly upload documents to Google's File Search Store
+- **Metadata Filtering**: Filter search results by custom metadata
+- **Grounding Support**: Get responses with citation metadata
+
+## Installation
+
+```bash
+pip install google-genai
+```
+
+## Configuration
+
+Set your Google API key as an environment variable:
+
+```bash
+export GOOGLE_API_KEY="your-api-key-here"
+```
+
+Or get one from: https://ai.google.dev/
+
+## Basic Usage
+
+```python
+from agno.agent import Agent
+from agno.knowledge.knowledge import Knowledge
+from agno.vectordb.gemini.gemini_file_search import GeminiFileSearch
+
+# Create Gemini File Search vector database
+vector_db = GeminiFileSearch(
+    file_search_store_name="my-knowledge-store",
+    model_name="gemini-2.5-flash-lite",
+    api_key="your-api-key",
+)
+
+# Create knowledge base
+knowledge = Knowledge(
+    name="My Knowledge Base",
+    vector_db=vector_db,
+)
+
+# Add documents
+knowledge.add_content(
+    name="MyDocument",
+    url="https://example.com/document.pdf",
+    metadata={"doc_type": "manual"},
+)
+
+# Create agent and query
+agent = Agent(knowledge=knowledge, search_knowledge=True)
+agent.print_response("What is covered in the document?")
+```
+
+## Examples
+
+- **[gemini_file_search.py](./gemini_file_search.py)** - Basic usage with Thai recipe knowledge base
+- **[async_gemini_file_search.py](./async_gemini_file_search.py)** - Async operations with Agno documentation
+- **[gemini_file_search_with_filters.py](./gemini_file_search_with_filters.py)** - Using metadata filters for refined search
+
+## Supported Operations
+
+| Operation | Supported | Notes |
+|-----------|-----------|-------|
+| `create()` | ✅ | Creates or gets existing File Search Store |
+| `insert()` | ✅ | Uploads documents to the store |
+| `search()` | ✅ | Semantic search with optional metadata filters |
+| `upsert()` | ✅ | Updates existing documents or inserts new ones |
+| `delete_by_name()` | ✅ | Delete documents by display name |
+| `delete_by_id()` | ✅ | Delete documents by ID |
+| `delete_by_content_id()` | ✅ | Delete documents by content ID |
+| `delete_by_metadata()` | ❌ | Not supported by Gemini File Search |
+| `update_metadata()` | ❌ | Not supported by Gemini File Search |
+
+## Important Notes
+
+1. **File Search Store**: Documents are organized in "File Search Stores" - named containers for your documents
+2. **Document Names**: Each document has both a system-generated `name` (ID) and a user-defined `display_name`
+3. **Operation Polling**: Document uploads are asynchronous; the library polls until completion
+4. **Metadata Limitations**: 
+   - Supports string, numeric, and float metadata values
+   - Metadata can be used for filtering during search
+   - Cannot update metadata after upload (must delete and re-upload)
+5. **Cost**: Check Google AI pricing for File Search Store usage
+
+## Model Options
+
+Gemini File Search supports various Gemini models:
+
+- `gemini-2.5-flash-lite` (default) - Fast and cost-effective
+- `gemini-2.5-flash` - Balanced performance
+- `gemini-2.0-flash` - High performance
+- `gemini-2.0-flash-exp` - Experimental features
+
+## API Reference
+
+See the [GeminiFileSearch documentation](../../../../libs/agno/agno/vectordb/gemini/gemini_file_search.py) for detailed API information.
+
+## Resources
+
+- [Google AI Gemini Docs](https://ai.google.dev/gemini-api/docs)
+- [File Search API](https://ai.google.dev/gemini-api/docs/file-search)
+- [Agno Documentation](https://docs.agno.com)

cookbook/knowledge/vector_db/gemini_file_search/__init__.py

@@ -0,0 +1 @@
+"""Gemini File Search cookbook examples."""

cookbook/knowledge/vector_db/gemini_file_search/async_gemini_file_search.py

@@ -0,0 +1,57 @@
+"""
+Async example of using Gemini File Search as a vector database.
+
+Requirements:
+- pip install google-genai
+- Set GOOGLE_API_KEY environment variable
+"""
+
+import asyncio
+from os import getenv
+
+from agno.agent import Agent
+from agno.knowledge.knowledge import Knowledge
+from agno.vectordb.gemini.gemini_file_search import GeminiFileSearch
+
+# Get API key from environment
+api_key = getenv("GOOGLE_API_KEY")
+
+# Initialize Gemini File Search
+vector_db = GeminiFileSearch(
+    file_search_store_name="agno-docs-store",
+    model_name="gemini-2.5-flash-lite",
+    api_key=api_key,
+)
+
+# Create knowledge base
+knowledge = Knowledge(
+    name="Agno Documentation",
+    description="Knowledge base with Agno documentation using Gemini File Search",
+    vector_db=vector_db,
+)
+
+# Create and use the agent
+agent = Agent(knowledge=knowledge, search_knowledge=True)
+
+
+async def main():
+    """Main async function."""
+    # Add content to the knowledge base
+    # Comment out after first run to avoid re-uploading
+    await knowledge.add_content_async(
+        name="AgnoIntroduction",
+        url="https://docs.agno.com/concepts/agents/introduction.md",
+        metadata={"doc_type": "documentation", "topic": "agents"},
+    )
+
+    # Query the knowledge base using async
+    await agent.aprint_response("What is the purpose of an Agno Agent?", markdown=True)
+
+    # Additional query
+    await agent.aprint_response(
+        "How do I create an agent with knowledge?", markdown=True
+    )
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

cookbook/knowledge/vector_db/gemini_file_search/gemini_file_search.py

@@ -0,0 +1,54 @@
+"""
+Basic example of using Gemini File Search as a vector database.
+
+Requirements:
+- pip install google-genai
+- Set GOOGLE_API_KEY environment variable
+"""
+
+import asyncio
+from os import getenv
+
+from agno.agent import Agent
+from agno.knowledge.knowledge import Knowledge
+from agno.vectordb.gemini.gemini_file_search import GeminiFileSearch
+
+# Get API key from environment
+api_key = getenv("GOOGLE_API_KEY")
+
+# Create Gemini File Search vector database
+vector_db = GeminiFileSearch(
+    file_search_store_name="thai-recipes-store",
+    model_name="gemini-2.5-flash-lite",
+    api_key=api_key,
+)
+
+# Create Knowledge Instance with Gemini File Search
+knowledge = Knowledge(
+    name="Thai Recipe Knowledge Base",
+    description="Agno 2.0 Knowledge Implementation with Gemini File Search",
+    vector_db=vector_db,
+)
+
+# Add content to the knowledge base
+# Note: This uploads documents to Gemini File Search Store
+asyncio.run(
+    knowledge.add_content_async(
+        name="Recipes",
+        url="https://agno-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf",
+        metadata={"doc_type": "recipe_book", "cuisine": "Thai"},
+    )
+)
+
+# Create and use the agent
+agent = Agent(knowledge=knowledge, search_knowledge=True)
+
+# Query the knowledge base
+agent.print_response("List down the ingredients to make Massaman Gai", markdown=True)
+
+# Delete operations examples
+# Delete by name
+vector_db.delete_by_name("Recipes")
+
+# Note: delete_by_metadata is not supported by Gemini File Search
+# You can only delete documents by name or ID

cookbook/knowledge/vector_db/gemini_file_search/gemini_file_search_with_filters.py

@@ -0,0 +1,83 @@
+"""
+Example of using Gemini File Search with metadata filters.
+
+Requirements:
+- pip install google-genai
+- Set GOOGLE_API_KEY environment variable
+"""
+
+import asyncio
+from os import getenv
+
+from agno.agent import Agent
+from agno.knowledge.knowledge import Knowledge
+from agno.vectordb.gemini.gemini_file_search import GeminiFileSearch
+
+# Get API key from environment
+api_key = getenv("GOOGLE_API_KEY")
+
+# Create Gemini File Search vector database
+vector_db = GeminiFileSearch(
+    file_search_store_name="multi-cuisine-recipes",
+    model_name="gemini-2.5-flash-lite",
+    api_key=api_key,
+)
+
+# Create Knowledge Instance
+knowledge = Knowledge(
+    name="Multi-Cuisine Recipe Knowledge Base",
+    description="Knowledge base with recipes from different cuisines",
+    vector_db=vector_db,
+)
+
+
+async def add_recipes():
+    """Add multiple recipe documents with different metadata."""
+    # Add Thai recipes
+    await knowledge.add_content_async(
+        name="ThaiRecipes",
+        url="https://agno-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf",
+        metadata={"cuisine": "Thai", "difficulty": "medium"},
+    )
+
+    # You can add more recipes from other sources with different metadata
+    # await knowledge.add_content_async(
+    #     name="ItalianRecipes",
+    #     content="Italian recipe content here...",
+    #     metadata={"cuisine": "Italian", "difficulty": "easy"},
+    # )
+
+
+async def main():
+    """Main function."""
+    # Add content (comment out after first run)
+    await add_recipes()
+
+    # Create agent
+    agent = Agent(knowledge=knowledge, search_knowledge=True)
+
+    # Query with specific filters
+    # Note: Gemini File Search supports metadata filtering in search
+    print("\n=== Searching for Thai recipes ===\n")
+    await agent.aprint_response(
+        "What are some popular Thai dishes?",
+        markdown=True,
+    )
+
+    # You can also search the vector database directly with filters
+    print("\n=== Direct vector DB search with filters ===\n")
+    results = vector_db.search(
+        query="coconut curry recipes",
+        limit=3,
+        filters={"cuisine": "Thai"},  # Filter by cuisine
+    )
+
+    for i, result in enumerate(results, 1):
+        print(f"\nResult {i}:")
+        print(f"Content: {result.content[:200]}...")
+        if result.meta_data:
+            print(f"Metadata: {result.meta_data}")
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

libs/agno/agno/vectordb/gemini/__init__.py

@@ -0,0 +1,5 @@
+from agno.vectordb.gemini.gemini_file_search import GeminiFileSearch
+
+__all__ = [
+    "GeminiFileSearch",
+]