[GH-ISSUE #268] Accessing documents in the pipelines container #126

New issue

Open

opened 2026-10-18 21:02:53 +00:00 by bsakaguchi · 1 comment

bsakaguchi commented

2026-10-18 21:02:53 +00:00

Owner

Originally created by @afonso-ccg on GitHub (Sep 23, 2024).
Original GitHub issue: https://github.com/open-webui/pipelines/issues/268

I am struggling to access the documents that I upload in the Open WebUI interface. How can I do it? Is there an API? In the Custom RAG example there is this piece of code:

self.documents = SimpleDirectoryReader("./data").load_data()

but for me that returns nothing, and I got to understand that as the pipeline is running on a separate container, the path that I’m trying to access in that piece of code is in the pipelines container. Is there any way to do this?

Originally created by @afonso-ccg on GitHub (Sep 23, 2024). Original GitHub issue: https://github.com/open-webui/pipelines/issues/268 I am struggling to access the documents that I upload in the Open WebUI interface. How can I do it? Is there an API? In the Custom RAG example there is this piece of code: self.documents = SimpleDirectoryReader("./data").load_data() but for me that returns nothing, and I got to understand that as the pipeline is running on a separate container, the path that I’m trying to access in that piece of code is in the pipelines container. Is there any way to do this?

bsakaguchi commented

2026-10-18 21:02:54 +00:00

Author

Owner

@jasgeo75 commented on GitHub (Dec 1, 2024):

There may be an easier way, but I've switched from using sqlite w/ webui.db (the default) to using postgres and qdrant. This allows for the separation needed for my pipeline to work. My script looks up the collections associated with the 'knowledge' in postgres and the I can feed those details in when searching in qdrant, but I do find the use of collections as they are currently implemented to be sub-optimal; it would be great to see multi-tenancy.

@jasgeo75 commented on GitHub (Dec 1, 2024): There may be an easier way, but I've switched from using sqlite w/ webui.db (the default) to using postgres and qdrant. This allows for the separation needed for my pipeline to work. My script looks up the collections associated with the 'knowledge' in postgres and the I can feed those details in when searching in qdrant, but I do find the use of collections as they are currently implemented to be sub-optimal; it would be great to see multi-tenancy.

bsakaguchi referenced this issue

2026-10-18 21:05:02 +00:00

[PR #126] [MERGED] fix: anthropic pipeline image upload #325