Vector Stores are not retrievable via list all

API Bugs
, ,
2

I called a few times and reloaded a few times. Seemed healthy on my few old examples, which is all I have.

Here’s some Python written to not fail on you.

  • You can put api_key="sk-proj-your_api_key_string" directly in code if local for assurance.
  • You can put in a loop and keep asking for your vector stores with a bit of time.sleep(5) to report on intermittent issues, or increase the default count.
  • Default results are max 20, and the API maximum return is 100. The playground pages 10 at a time.
import os
import json
import gzip
import zlib
import urllib.request
import urllib.parse

def decode_content(data, encoding):
    """
    Decompress response body according to Content-Encoding header.
    Supports: gzip, deflate. Other encoding values are returned as-is.
    """
    if not encoding:
        return data
    enc = encoding.lower().strip()
    if enc == "gzip":
        return gzip.decompress(data)
    if enc == "deflate":
        return zlib.decompress(data)
    return data

def list_vector_store_ids(
        limit=20,
        after=None,
        before=None,
        order="desc"
    ):
    """
    Fetch vector store entries from OpenAI and return a list of IDs.

    Parameters:
      limit   - max number of items to return (1–100, default 20)
      after   - cursor ID to fetch items after this ID
      before  - cursor ID to fetch items before this ID
      order   - "asc" or "desc" sort on created_at (default "desc")

    Returns:
      A list of string IDs for each vector_store in the response.
    """
    api_key = os.getenv("OPENAI_API_KEY")
    if not api_key:
        raise EnvironmentError(
            "OPENAI_API_KEY environment variable is not set"
        )

    # build query parameters
    params = {}
    if limit is not None:
        params["limit"] = str(limit)
    if after:
        params["after"] = after
    if before:
        params["before"] = before
    if order:
        params["order"] = order
    qs = urllib.parse.urlencode(params)

    # assemble the full URL
    url = (
        "https://api.openai.com/v1/vector_stores"
        "?" + qs
    )

    headers = {
        "Accept": "*/*",
        "Accept-Encoding": "gzip, deflate",
        "Accept-Language": "en-US,en;q=0.5",
        "Authorization": "Bearer " + api_key,
        "Cache-Control": "no-cache",
        "Connection": "keep-alive",
        "Host": "api.openai.com",
        "OpenAI-Beta": "assistants=v2",
        "Pragma": "no-cache",
        "Priority": "u=0",
        "User-Agent": (
            "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:143.0) "
            "Gecko/20100101 Firefox/143.0"
        ),
    }

    req = urllib.request.Request(url, headers=headers, method="GET")
    with urllib.request.urlopen(req) as resp:
        raw = resp.read()
        encoding = resp.getheader("Content-Encoding")
        body = decode_content(raw, encoding)
        data = json.loads(body.decode("utf-8"))

    ids = []
    for item in data.get("data", []):
        vid = item.get("id")
        if vid:
            ids.append(vid)
    return ids

if __name__ == "__main__":
    # Example usage: fetch up to 10 most recent stores
    vector_ids = list_vector_store_ids(limit=10)
    for vid in vector_ids:
        print(vid)

'''
The vector store list call returns a list of items. If there was only one:

{
  "object": "list",
  "data": [
    {
      "id": "vs_abcdef1234567890",
      "object": "vector_store",
      "created_at": 1700000000,
      "name": "example vector store",
      "description": null,
      "usage_bytes": 2048,
      "file_counts": {
        "in_progress": 0,
        "completed": 1,
        "failed": 0,
        "cancelled": 0,
        "total": 1
      },
      "status": "completed",
      "expires_after": null,
      "expires_at": null,
      "last_active_at": 1700000000,
      "metadata": {}
    }
  ],
  "first_id": "vs_abcdef1234567890",
  "last_id": "vs_abcdef1234567890",
  "has_more": false
}
'''

The API endpoint itself has rate limits for calls that are non-consuming of AI resources. They are pretty high but reachable.