feat(collection-browse): tabular collection browse view behind feature flag

[orhanrauf]

Summary

• Adds an unranked, paginated tabular browse view of a collection alongside the existing search experience, gated on a new COLLECTION_BROWSE org feature flag.
• Backend reuses VespaVectorDB.filter_search() + count() and forces chunk_index = 0 so each source entity shows up as exactly one row. New BrowseService is exposed through the DI container and wired into /collections/{readable_id}/search/browse.
• Frontend ships a BrowseTable component (rows, sync/entity-type filters, debounced substring name search, offset/limit pagination, row drawer, CSV/JSON export) wired into CollectionDetailView as a Tabs sibling to Search.
• Name search uses a dedicated regex path on the vector-db protocol (name matches "(?i).*<escaped>.*") rather than contains, because Vespa's contains operator on attribute fields is whole-token match — typing "Qui" wasn't matching "Quick process to debug" via the existing FilterCondition translator.

CI status (verified locally on changed lines)

• ruff check + ruff-format: 100% clean (diff-quality)
• mypy: 100% clean on changed lines (diff-quality) — full-repo baseline is unchanged
• import-linter: 0 broken
• backend unit tests: 4201 passed (added FakeBrowseService + threaded it through the test_container fixture so existing tests don't break on the new required Container.browse_service field)
• ESLint: clean
• Frontend production build: clean

Test plan

• Enable the COLLECTION_BROWSE feature flag on an org and open a collection — verify the Browse tab appears next to Search.
• Verify pagination, sync filter, entity-type filter all work independently and in combination.
• Type into the name search (e.g. partial token like "Qui") — confirm reactive results, that special characters (., +, () don't break the query, and that aborts work when typing fast.
• Export current page → CSV and JSON, then export "all matching" → confirm capped at 1000 rows.
• Click a row, confirm the drawer opens and entity_id copy-to-clipboard works.
• Disable the flag → Browse tab disappears and /search/browse returns 403/feature-disabled.

🤖 Generated with Claude Code

---

Summary by cubic

Adds a feature-flagged tabular browse view for collections with a new POST /collections/{readable_id}/search/browse endpoint and a Browse tab in the UI. It lists entities unranked with pagination, filters, case-insensitive substring name search (properly escaped), and CSV/JSON export.

• New Features

  • Backend: POST /collections/{readable_id}/search/browse gated by FeatureFlag.COLLECTION_BROWSE; uses VectorDBProtocol.filter_search() + count() in parallel with chunk_index = 0; supports name_query via a regex matches clause in VespaVectorDB (case-insensitive substring with escaping); adds BrowseService, DI wiring, and BrowseRequest/BrowseResponse.
  • Frontend: BrowseTable with source filter, debounced name search, offset/limit pagination, row drawer, open-in-source link, and CSV/JSON export (cap 1000 for “all matching”); added Tabs in CollectionDetailView; new FeatureFlags.COLLECTION_BROWSE.

• Bug Fixes

  • Moved route to /search/browse for consistency with other tiers.
  • Escaped double quotes in the YQL name-substring regex to prevent parse errors and injection; added tests for the clause, endpoint flag gate (404 when disabled), and service behavior.
  • Input and request bounds: sync_ids/entity_types capped at 100; name_query requires ≥2 chars (frontend hints and only sends when long enough); export fetches in 200-row chunks to respect the backend page limit.

^{Written for commit bcbf327. Summary will update on new commits.}

[cubic-dev-ai]

2 issues found across 16 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="backend/airweave/domains/search/adapters/vector_db/vespa_client.py">

<violation number="1" location="backend/airweave/domains/search/adapters/vector_db/vespa_client.py:316">
P1: Escape double quotes in `name_substring` before interpolating into the double-quoted YQL regex literal.</violation>
</file>

<file name="backend/airweave/api/v1/endpoints/search.py">

<violation number="1" location="backend/airweave/api/v1/endpoints/search.py:86">
P1: The browse endpoint path is missing the `/search` segment, so it won’t be reachable at the documented `/.../search/browse` URL.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}