You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is your feature request related to a problem? Please describe.
Codebase Indexing is difficult to setup with multiple dependencies to install. competition comes with it built in.
Describe the solution you'd like
Add semble code base indexing, in addition to the current solutions. Should be as easy as enabling it with a checkmark. Consider making it default if nothing is configured
Describe alternatives you've considered
Alternatives already exist
My current setup is with the OpenAI Compatible provider, using the text-embedding-3-large model. But I also have to run Qdrant which is a bit of a faff to get running locally (as I don't have an option of using it remotely). Setting up semble manually and providing it through an MCP server to Zoo is perhaps even trickier.
The point is, it's fragile.
A checkbox to enable semble, is foolproof. And I agree it should be the default. And I think the Zoo devs should consider deprecating Qdrant-based indexing when semble is implemented and working. Just to reduce any potential confusion.
Could we also incorporate some kind of pruning task? I find the current qdrant setup just keeps loading new files into the db and never prunes out old ones, so eventually qdrant reindexing blows up docker.
Update I have a working version locally it integrates with semble installed on your system. The issue is the current released semble binary indexes every time you search which is horribly inefficient. They recenly remedied that by adding an index command so the index is cached. I will wait to integrate with that
Activity
marty-a11y commentedon May 20, 2026
My current setup is with the OpenAI Compatible provider, using the
text-embedding-3-largemodel. But I also have to run Qdrant which is a bit of a faff to get running locally (as I don't have an option of using it remotely). Setting up semble manually and providing it through an MCP server to Zoo is perhaps even trickier.The point is, it's fragile.
A checkbox to enable semble, is foolproof. And I agree it should be the default. And I think the Zoo devs should consider deprecating Qdrant-based indexing when semble is implemented and working. Just to reduce any potential confusion.
navedmerchant commentedon May 20, 2026
I am taking this on
edelauna commentedon May 21, 2026
Could we also incorporate some kind of pruning task? I find the current qdrant setup just keeps loading new files into the db and never prunes out old ones, so eventually qdrant reindexing blows up docker.
navedmerchant commentedon May 26, 2026
Update I have a working version locally it integrates with semble installed on your system. The issue is the current released semble binary indexes every time you search which is horribly inefficient. They recenly remedied that by adding an index command so the index is cached. I will wait to integrate with that
kazenshi commentedon May 27, 2026
@navedmerchant Looks like a new release is out adding what you mentioned: https://github.com/MinishLab/semble/releases/tag/v0.3.0