Summary

Promote memory_dirs management from inside the Config tab (Indexing card) to its own section of the existing Sources tab. The current location works but the widget fights for vertical space inside a config card, truncates long paths, and conceptually mixes "action-based immediate persist" semantics with the rest of the "batched Save" config fields.

Context

Shipped in #295: a categorized + editable memory_dirs widget with per-item remove/reindex, per-group reindex, top-level reindex-all, inline add input, and per-dir "(N chunks) / (not indexed) / missing" status badges fed by GET /api/memory-dirs/status.

Observed during smoke on a real post-v0.1.12 config (29 dirs, 4 categories):

• Paths truncate (/Users/pdstudio/memo…) because the widget sits in a 2-column config-table cell.
• The widget is an "immediate persist" island inside a Save/dirty section — users see a (disabled, since _NO_RESET_FIELDS) reset button column and no Save button for it, which is inconsistent with the rest of the card.
• Sources tab already has per-file stats UI, so memory_dirs fits there conceptually: Sources = where memory comes from; memory_dirs = the roots of those sources.

Proposed design

Layout

• Move memory_dirs widget out of Config tab → Sources tab, as a top panel above the existing per-file stats list.
• Full-width layout (no 2-column grid): each dir row gets enough space for the absolute path + badges + actions without truncation.
• Keep the 4-category collapsible grouping (User / Claude projects / Claude plans / Codex) that shipped in fix: memory_dirs editable widget with per-dir index status #295.
• Existing Sources tab content (per-file stats, source list) stays below; filter it by the currently-expanded dir group, or leave cross-dir for now.

Config tab

• Leave a breadcrumb in the Indexing card ("Memory Dirs are managed in Sources tab") so users who look for it in the old location get redirected.
• Keep env var hint + chunk settings where they are.

Per-dir stats enrichment (optional, follow-on within same issue)

Sources tab's additional real estate lets us surface richer state per dir beyond the minimal badges shipped in #295:

• Last indexed timestamp (already in DB via MAX(updated_at))
• File count + chunk count + avg chunks/file
• Click → drill into the existing per-file stats filtered to that dir
• Maybe a sparkline or badge for "stale files" (mtime > last_indexed) if we want to nudge reindex

Scope boundaries

• Keep the same set of API endpoints (GET /api/memory-dirs/status, POST /api/memory-dirs/{add,remove}, POST /api/{index,reindex}). No backend changes expected beyond maybe extending memory_dir_stats with last_updated.
• Don't touch the categorization logic (_detect_provider_dirs in config.py) — it's the server-side source of truth for client-side bucketing.
• i18n: keys migrate from settings.memory_dirs.* to sources.memory_dirs.* (or keep as-is with cross-refs). Preserve en/ko parity (existing test_i18n.py guards).

Why not in #295

#295 was already bundling two concerns (UI readability + missing-index bug). Adding a tab-level reshuffle would have doubled the review surface and risked drift from the core "make memory_dirs editable + status-visible" goal. Shipping the widget first in its current location let us validate the primitives (badge + action-based persist) before redesigning the container.

Related

• Ships on top of fix: memory_dirs editable widget with per-dir index status #295
• Referenced in memo feedback_pullbased_vs_startup_scan.md — the per-dir status primitive is the reusable piece