{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreigdkq3wabsyrn3zf2urywtcn3crc3lnyliavxbzmpazv65a7ss7gi",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mkuhpjlkjkb2"
  },
  "path": "/t/bug-vector-store-status-completed-does-not-guarantee-searchability-file-search-returns-empty-results-silently/1374471#post_3",
  "publishedAt": "2026-05-02T09:57:23.000Z",
  "site": "https://community.openai.com",
  "textContent": "Facing a similar issue when using `file_search` in the Assistants API (semantic search on the retrieval API may also be affected). I’ve implemented a fallback semantic search function and added instructions to use this fallback function if `file_search` doesn’t find any results. The fallback function takes a `query` and uses OpenAI’s retrieval API to find relevant chunks from my vector store and include them in the current context.\n\nWhenever `file_search` doesn’t find any results, the fallback function is called immediately but it also doesn’t find any relevant chunks. However, if I hit the retrieval API manually via curl, with the exact same query, configs and everything else, searching the same vector store, I get chunks (most of the time). The search fails approx once every 50 requests, and I haven’t been able to reliably reproduce the failure.\n\nSeems like an issue with the vector stores availability, and not the `file_search` tool specifically.",
  "title": "Bug: Vector store status: completed does not guarantee searchability - file_search returns empty results silently"
}