feat: remove s3 uploads from local storage

[louis-fiori]

Check Existing Issues

• I have searched the existing issues and discussions.

Problem Description

We’re currently using the S3 storage provider in OpenWebUI to comply with internal data retention and compliance requirements. Specifically, all uploaded files must reside exclusively in S3.

However, we noticed that while the S3 provider does upload the file to S3 successfully, the file also remains on the local container’s storage. This creates a compliance issue for us, as files must not persist outside of S3.

Desired Solution you'd like

We propose adding a configuration option to automatically delete the local file after a successful upload to S3.

The update could be made in the storage/provider.py file, within the upload_file method of the S3StorageProvider.

def upload_file(self, file_path: str, file_name: str) -> str:
    ...
    self.s3.upload_file(file_path, self.bucket_name, s3_key)
    
    if self.delete_after_upload:
        os.remove(file_path)

This could be controlled via an environment variable (e.g., S3_DELETE_LOCAL_AFTER_UPLOAD=True), or a config flag.

Alternatives Considered

We are currently manually deleting the files after upload via a patch, but this is not ideal or maintainable long-term.

Additional Context

No response

[tjbck]

PR Welcome, however, just so you know when the file is needed it'll be downloaded to the instances.

[erhhung]

• Does this "auto-download from cloud storage if the local file isn't found" behavior already exist?
• Where does the backend expect the file to reside? always under /app/backend/data/uploads/?
• If it has to re-download from cloud storage, does it store those files under /app/backend/data/uploads/ or /app/backend/data/cache/?
• Is there a mechanism to auto-purge cached content when space is low? My deployment is on K8s, and I created a PVC specifically for /app/backend/data/cache/, so I hope the file caching mechanism can work within the alloted storage. Since /app/backend/data is automatically mounted as an emptyDir, I also expect /app/backend/data/uploads/ not to get filled up because I'm also using S3 for permanent storage.

[schneider-f]

Also interested in some clarification around this behavior.

[oguzhanogreden]

I'm interested in digging into this and happy to help move it forward. We'd very much like to not store data on our openwebui instances.

Step 1 — Identify where raw file data is actually needed

Some cases will require a local copy. For those, we should define what happens after use. Two main options:

• Delete right after use → simpler, but may cause repeated downloads and higher network costs.
• Delete after a configurable delay → fewer downloads, but adds complexity.

For everything else, we can stream directly from S3. File downloads by users fit here, and maybe some integrations too.

Step 2 — Handle these case-by-case

Once we’ve mapped the use cases, we can work toward a reference implementation and reuse it where needed.

---

Open to feedback on this approach or alternative angles to explore. @tjbck , I'm especially interested in hearing if you have any strong opinion e.g. to avoid streaming directly for downloads, etc.