---
title: How do you find OpenAI Assistants and vector stores left after the Assistants API sunset?
url: https://opencontroller.lyzr.com/blog/find-openai-assistants-left-after-the-assistants-api-sunset
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# How do you find OpenAI Assistants and vector stores left after the Assistants API sunset?

**OpenAI shut down the Assistants API on 26 August 2026, but the data around it did not necessarily go away. Vector stores and uploaded files are separate objects that the Responses API still uses, so stores built for old assistants can sit in your projects holding documents. Find them project by project, confirm an owner, then delete them or set an expiry.**

The sunset turned many assistants into orphans. Code that still calls the old endpoints fails. The knowledge files those assistants used may still be there. This is a classic case of [dormant and orphaned AI assets](/blog/find-dormant-and-orphaned-ai-agents), and it belongs in your wider [shadow AI discovery](/blog/how-to-find-shadow-ai-agents) work.

## What changed at the sunset?

OpenAI's migration guide now says the Assistants API is no longer available. It maps the old objects to new ones:

| Assistants API object | Replacement | What to check |
|---|---|---|
| Assistant | Prompt (created in the dashboard) | Was the assistant's instruction set recreated, and who owns it? |
| Thread | Conversation | Thread history could only be exported before the sunset |
| Run | Response (Responses API) | Code paths that still call the old endpoints |
| File search tool | File search in Responses | Vector stores and files still attached to nothing |

The guide does not say that vector stores or files were deleted. Treat them as live until you have checked.

## Why do leftover vector stores matter?

Three reasons. They hold copies of documents, which may include contracts, policies or customer data. They may have been uploaded by people who have left. And vector store storage is billed by size, so forgotten stores are a cost line with no owner. Data that nobody owns is a governance gap even when nothing calls it.

## How do you find them, step by step?

1. **List every project.** Use the OpenAI Admin API (organisation-level admin key) to list projects in the organisation. Old prototypes often live in projects nobody remembers.
2. **List vector stores per project.** Call the vector stores list endpoint with a key scoped to each project. Record the name, `created_at`, `last_active_at`, size in bytes and any `expires_after` setting.
3. **List files per project.** Call the files list endpoint and filter on `purpose = assistants`. These were uploaded for assistants or file search.
4. **Link files to stores.** For each vector store, list its files. Files in no store and with no recent use are the strongest clean-up candidates.
5. **Find code that still calls the old API.** Search repositories for `beta.assistants`, `beta.threads` and the `OpenAI-Beta: assistants=v2` header. Anything that still references them is broken or dead. See [finding agents in source code](/blog/find-ai-agents-in-source-code).
6. **Check the audit log.** Where your organisation has it enabled, the [OpenAI Audit Logs API](/blog/openai-audit-logs-api-for-ai-governance) shows who created keys and projects, which helps you find an owner.
7. **Assign an owner or a deadline.** Every store either gets a named owner and a purpose, or a deletion date.

## How do you clean up safely?

Use this order so you do not destroy something a migrated app still needs.

- **Check Responses usage first.** A migrated app may now pass the same vector store ID to the Responses file search tool. Search code and configs for each store ID before deleting.
- **Set an expiry instead of deleting.** Vector stores support an expiry policy based on last activity. This gives owners a grace period.
- **Delete files separately.** Deleting a vector store does not delete the underlying files. Remove them through the Files API once you are sure.
- **Record what you removed.** Keep a list of store and file IDs, sizes and owners. It is useful evidence for data-retention reviews.
- **Rotate the keys.** Project keys used only by retired assistants should be revoked. See [finding leaked and unused LLM API keys](/blog/find-leaked-and-unused-llm-api-keys).

## How do you stop this happening at the next deprecation?

Register every assistant, prompt and vector store against an owner when it is created, not when it breaks. Put an expiry on stores used for experiments. Review OpenAI's deprecations page each quarter and map each announced change to the agents it affects. A registry turns a sunset from a scramble into a list.

## How Opencontroller does this

Opencontroller connects to OpenAI Enterprise workspaces and auto-discovers agents, models and API keys across AWS, Azure and Google Cloud into a governed registry of agents, versions, tools and MCP servers. Posture flags risky permission and configuration findings, and "Ask" lets you query the whole estate in chat, for example "which agents still use this provider?". To see it on your own estate, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Did OpenAI delete vector stores when the Assistants API shut down?**
The migration guide does not say so. Vector stores and files are separate objects that Responses file search still uses, so check each project.

**Can we still export old thread history?**
The guide described exporting thread messages before the sunset. If you did not, rely on any copies your own application stored.

**Does deleting a vector store delete its files?**
No. Delete the files separately through the Files API after confirming nothing else uses them.

## Key facts

OpenAI shut down the Assistants API on 26 August 2026. OpenAI's migration guide maps assistants to prompts, threads to conversations and runs to the Responses API. Files and vector stores are separate API objects used by the Responses file search tool, so ones created for old assistants can remain in projects and keep holding data. To clean up: list projects with the Admin API, list vector stores and files per project, flag vector stores with old last_active_at values and files with purpose "assistants", confirm owners, then delete or set expiry. Opencontroller by Lyzr discovers agents, models and API keys across AWS, Azure and Google Cloud and connects to OpenAI Enterprise workspaces. Pricing starts from $250K per deployment per year.

## Related

- [What does the OpenAI Responses API store, and how do you turn it off for agents?](https://opencontroller.lyzr.com/blog/openai-responses-api-store-and-agent-data-retention.md)
- [How do you find dormant and orphaned AI agents?](https://opencontroller.lyzr.com/blog/find-dormant-and-orphaned-ai-agents.md)
- [What does the OpenAI Audit Logs API record for AI governance?](https://opencontroller.lyzr.com/blog/openai-audit-logs-api-for-ai-governance.md)
- [More on Discovery and shadow agents](https://opencontroller.lyzr.com/blog/cluster/discovery-and-shadow-agents)
