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
{{ message }}
Repository navigation
MANUAL RUN: turn on search across repos (CODESPLITTERS_VECTORS=1) and backfill #704
The vector index across repos (#636) ships off. Turning it on takes the owner's deploy:
Add CODESPLITTERS_VECTORS=1 to the release script, beside CODESPLITTERS_DEPLOYS=1 CODESPLITTERS_AGENTS=1.
plan should show new: Vectorize index codesplitters-vectors (384 dims, cosine), its two metadata indexes (repo, pub), and the Worker's VECTORS and AI bindings. Migration 0052 adds repo_search.stale.
Deploy. If wrangler's login lacks ai:write, log in again (rustybuns asks for it now).
Backfill: POST /api/admin/scale/search-embed as an admin until left is 0, or let the hourly cron do 200 files an hour.
Check: a search across repos with no repo= finds a repo past the 50 most recently committed; the scale sideboard's search card reads "Yes".
Cost: the backfill is the spend, about one embedding per 40 lines of every committed file. Workers AI's free 10k neurons a day cover ~5M tokens; Vectorize on Workers Paid includes 10M stored dimensions (~26k chunks).
Activity
QuinnsCode commentedon Oct 11, 2026
Tracked in #609, the one hands-on checklist. The steps here still apply; tick them there.
🤖 Generated with Claude Code