Source: File search tool (OpenAI guide)

OpenAI's hosted file search tool: vector stores, file_search_call output, file_citation annotations, max_num_results and metadata filters.

Contents
  • What: official API documentation page
  • Publisher: OpenAI
  • Date: living document; read 2026-11-14
  • Original: File search

Summary

Describes OpenAI’s hosted retrieval tool: upload files, attach them to a vector store, and let the model search it during a response.

Key claims

  • Hosted tool combining semantic and keyword search; the model decides when to call it.
  • Setup: upload with purpose: "assistants", create a vector store, add the file, poll until completed.
  • Tool definition {"type": "file_search", "vector_store_ids": [...]}; output holds a file_search_call item and a message with file_citation annotations (file_id, filename).
  • max_num_results trades quality against tokens and latency; include: ["file_search_call.results"] returns the raw hits (otherwise null); attribute filters narrow the search.
  • Supports PDF, DOCX, PPTX, Markdown, JSON, HTML, code and text files (text must be UTF-8/UTF-16/ASCII).
  • Available in Responses, Chat Completions and Assistants; 500 RPM on Build, 1,000 RPM on Launch and Grow.

Feeds

OpenAI Responses API

Suggest changes →