Skip to main content

Overview

The File Search block searches the files uploaded to your workspace and returns an AI-generated answer grounded in that content. Use it to answer questions from internal documents, pull facts from PDFs, or feed source-backed context into later steps. Every answer includes citations to the files it came from.

Configuration

Search Query (Required)

The question or topic to search for in your files. Example:
You can reference placeholders from previous steps: {{step_1.output}}

Max Number of Results

Maximum number of file matches to consider when building the answer. Default: 5 Range: 1–100

Context Chunk Radius

Expands page-level fallback context by including this many chunks before and after each match. Default: 0 Range: 0–10 Block-aware results are returned directly when available; this setting only affects page-level fallback.

Folder (Optional)

Restrict the search to a specific folder in your file drive. Leave empty to search across all files you can access.

Filters (Optional)

Add rules to narrow which files are searched. When you add more than one rule, choose And or Or to combine them. Filter fields:

Distinct files only

Return at most one result per file instead of multiple hits from the same document. Default: On

Output Format

Choose how the result is returned. Default: Plain Text

Error Handling

Define what happens if the block fails. Options:
  1. Terminate Workflow: Stop execution immediately.
  2. Continue Execution: Proceed to the next step despite the error.

Output Data

Plain Text format

Returns a single text string: the grounded answer, followed by a References: list and Supporting excerpts: when available. Accessing the answer: {{step_1.output}}

Structured (JSON) format

answer

The grounded answer to the query. Example: "Customers may return items within 30 days of delivery for a full refund."

query

The query that was searched.

citations

An array of citation objects referencing the files used.

hits

An array of supporting matches, each with file name, content excerpt, and score. Accessing fields: {{step_1.output.answer}}, {{step_1.output.citations[0]}}

Sample Output

Structured (JSON) format:

Best Practices

  • Write the query as a clear question for the most accurate grounded answer.
  • Use a folder or filters to scope large workspaces and improve relevance.
  • Choose Structured (JSON) when later steps need the citations or hits separately.
  • Choose Plain Text when feeding the answer directly into an LLM or document.
  • Keep “Distinct files only” on to avoid duplicate hits from one document.

Common Use Cases


What’s Next