[GH-ISSUE #3215] [FEAT]: Per-user customization in hosted (docker mode) for selecting preferred embedding model #2065

Closed
opened 2026-02-22 18:28:00 -05:00 by yindo · 1 comment
Owner

Originally created by @emad-qadri on GitHub (Feb 14, 2025).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/3215

What would you like to see?

Per-user customization in hosted (docker mode) for selecting a preferred embedding model. The use should be able to select an embedding model of preference for their specific use case.

Originally created by @emad-qadri on GitHub (Feb 14, 2025). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/3215 ### What would you like to see? Per-user customization in hosted (docker mode) for selecting a preferred embedding model. The use should be able to select an embedding model of preference for their specific use case.
yindo added the enhancementfeature request labels 2026-02-22 18:28:00 -05:00
yindo closed this issue 2026-02-22 18:28:00 -05:00
Author
Owner

@timothycarambat commented on GitHub (Feb 14, 2025):

The use should be able to select an embedding model of preference for their specific use case.

Unfortunately, we will not be supporting this use case. Since vector databases for the most part cannot support multi-dimension embeddings and typically models do not output the exact same dimensions. This is why the embedder is only set at the system level - as is the vector database.

These two elements are both very inflexible in their implementation and use. Otherwise, if we allowed this we will have 1000+ issues of people breaking their own instances due to dimensionality mismatching.

I do understand the use case requested, but it would currently require a massive overhaul of the system for what could be a minimal gain vs 1 good single embedding model :/

@timothycarambat commented on GitHub (Feb 14, 2025): > The use should be able to select an embedding model of preference for their specific use case. Unfortunately, we will not be supporting this use case. Since vector databases for the most part cannot support multi-dimension embeddings and typically models do not output the exact same dimensions. This is why the embedder is only set at the system level - as is the vector database. These two elements are both very _inflexible_ in their implementation and use. Otherwise, if we allowed this we will have 1000+ issues of people breaking their own instances due to dimensionality mismatching. I do understand the use case requested, but it would currently require a massive overhaul of the system for what could be a minimal gain vs 1 good single embedding model :/
yindo changed title from [FEAT]: Per-user customization in hosted (docker mode) for selecting preferred embedding model to [GH-ISSUE #3215] [FEAT]: Per-user customization in hosted (docker mode) for selecting preferred embedding model 2026-06-05 14:44:20 -04:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#2065