Import uses wrong namespace for embedder vectors — search returns no results #51

New issue

Closed

opened 2026-02-13 14:45:29 +00:00 by mik-tf · 0 comments

mik-tf commented

2026-02-13 14:45:29 +00:00

Owner

Problem

When importing a git repo via the Import UI, two different namespaces are used:

Library namespace (derived from git URL org) — e.g. znzcybercity — used for the library page, search UI, and disk structure at ~/hero/var/books/znzcybercity/
Embedder namespace (from the form's namespace field or collection name) — e.g. cybercity — used for vector storage in hero_embedder

The search page queries the library namespace (znzcybercity) which has 0 documents, while the vectors sit in cybercity with 2005 documents. Result: "No results found" for every query on a freshly imported library.

Steps to Reproduce

Delete znzcybercity from the system completely
Import via UI: git URL https://forge.ourworld.tf/znzcybercity/docs_znzcybercity
Go to /library/znzcybercity and search anything
"No results found" — vectors are in namespace cybercity, not znzcybercity

Expected Behavior

The import pipeline should use a single consistent namespace for both the library directory structure and the embedder vector storage.

Solution

Ensure import_collection_pipeline() uses the same namespace for:

scan_collections() namespace assignment
VectorStoreConfig namespace for embedder uploads
Library config and disk directory creation

The library namespace (derived from git org) should be the canonical one used everywhere.

## Problem When importing a git repo via the Import UI, two different namespaces are used: 1. **Library namespace** (derived from git URL org) — e.g. `znzcybercity` — used for the library page, search UI, and disk structure at `~/hero/var/books/znzcybercity/` 2. **Embedder namespace** (from the form's namespace field or collection name) — e.g. `cybercity` — used for vector storage in hero_embedder The search page queries the library namespace (`znzcybercity`) which has 0 documents, while the vectors sit in `cybercity` with 2005 documents. Result: **"No results found"** for every query on a freshly imported library. ## Steps to Reproduce 1. Delete znzcybercity from the system completely 2. Import via UI: git URL `https://forge.ourworld.tf/znzcybercity/docs_znzcybercity` 3. Go to `/library/znzcybercity` and search anything 4. "No results found" — vectors are in namespace `cybercity`, not `znzcybercity` ## Expected Behavior The import pipeline should use a single consistent namespace for both the library directory structure and the embedder vector storage. ## Solution Ensure `import_collection_pipeline()` uses the same namespace for: - `scan_collections()` namespace assignment - `VectorStoreConfig` namespace for embedder uploads - Library config and disk directory creation The library namespace (derived from git org) should be the canonical one used everywhere.