Google File Search icon

Google File Search

Interact with Google File Search API for RAG operations

Actions11

Query → Search

AI-generated

Summary

Execute a semantic search query on specified Google File Search stores using a selected Gemini model, optionally filtering results by metadata and returning either plain text or structured JSON output.

Inputs

  • model — Select the Gemini model to generate the search response, defaulting to Gemini 3 Flash Preview.
  • storeNames (required) — Choose one or more File Search stores by name or ID to be searched.
  • query (required) — Provide the search text query to execute against the selected stores; can be set dynamically from input data fields such as 'query', 'prompt', 'chatInput', or 'input'.
  • metadataFilter — Optional AIP-160-like syntax string to filter the search query results by metadata.
  • structuredOutput — Toggle to receive a response formatted as a structured JSON object matching a user-defined JSON schema (requires Gemini 3 or higher models).
  • jsonSchema — Define the JSON Schema when using structured output to specify the structure of the returned JSON response.
  • additionalFields — Optional parameters including whether to include source citations in the response, maximum tokens for the output, system prompt instructions for the model, and temperature controlling response randomness.

Output shape

a JSON object representing the search results with either a plain text answer or a structured JSON response matching the provided schema.

If structured output is enabled and the response can be parsed as JSON, the node returns an object under 'data' according to the JSON schema; optionally includes 'groundingMetadata' with source citation data if enabled. If parsing fails, the response includes raw text and a parse error flag. If structured output is disabled, returns plain text with optional model metadata. Errors from the API are surfaced or handled according to the node's continueOnFail setting.

Examples

Example 1: Search Google File Search stores for a query and receive a plain text answer

model: Gemini 3 Flash Preview; storeNames: ['fileSearchStores/store1', 'fileSearchStores/store2']; query: 'What are the main findings in the report?'; structuredOutput: false; includeGrounding: true.

Example 2: Search with structured JSON output

model: Gemini 3 Pro Preview; storeNames: ['fileSearchStores/123']; query: 'Summarize the document'; structuredOutput: true; jsonSchema: specifies an object with 'summary' (string) and 'references' (array of strings); includeGrounding: false.

Links

Discussion