---

name: filesearch version: 1.0.0 description: Gemini File Search (RAG) - semantic search over your documents

File Search Skill (Gemini RAG)

Enable Retrieval Augmented Generation (RAG) using Gemini's File Search API. Upload documents, create knowledge stores, and query with semantic search.

Features

• Semantic Search: Find relevant info by meaning, not just keywords
• Auto Chunking: Documents automatically split and indexed
• Free Storage: No cost for storage or query-time embeddings
• Citations: Responses include source references
• Structured Output: Get typed responses (Gemini 3+)

Capabilities

Supported File Types

• Documents: PDF, Word, Excel, PowerPoint
• Code: Python, JS, TS, Java, Go, Rust, etc.
• Text: Markdown, TXT, CSV, JSON, XML, YAML
• Web: HTML, CSS

Usage Examples

Create Store and Upload Files

Copyfrom rhea_noir.skills.filesearch.actions import skill as fs

# Create a knowledge store
store = fs.create_store("rhea-knowledge-base")

# Upload documents
fs.upload_file(store, "./docs/architecture.md")
fs.upload_file(store, "./docs/api-reference.pdf")

Query with Semantic Search

Copyresult = fs.query(
    store_name=store,
    question="How does the authentication system work?",
    model="gemini-3-flash-preview"
)
print(result["answer"])
print(result["citations"])

Add Metadata for Filtering

Copyfs.upload_file(
    store,
    "./books/i-claudius.txt",
    metadata={"author": "Robert Graves", "year": 1934}
)

# Query with filter
result = fs.query(
    store_name=store,
    question="What happened to Claudius?",
    metadata_filter='author="Robert Graves"'
)

Structured Output

Copyfrom pydantic import BaseModel

class Summary(BaseModel):
    title: str
    key_points: list[str]
    
result = fs.query_structured(
    store_name=store,
    question="Summarize the main architecture",
    response_schema=Summary
)

Pricing

• Indexing: $0.15 per 1M tokens (one-time)
• Storage: FREE
• Query embeddings: FREE
• Retrieved tokens: Normal Gemini pricing

Limits

[!NOTE] Keep stores under 20 GB for optimal retrieval latency.