feat(KnowledgeBase): add document viewer, download, metadata, and sorting

[bravosierra99]

Adds view, download, and metadata improvements to the Knowledge Base stored files table.

Changes

Backend (rag_controller.ts, rag_service.ts, rag.ts, routes.ts, types/rag.ts)

• getStoredFiles now returns StoredFile[] with fileName, size, uploadedAt, and isUserUpload instead of string[]
• New GET /api/rag/files/content — serves text file contents for inline viewing; restricted to text extensions, path-traversal safe (app directory only)
• New GET /api/rag/files/download — serves file as attachment; same path guard

Frontend (KnowledgeBaseModal.tsx, api.ts)

• Upload date and file size columns in the stored files table, both sortable by clicking the column header
• Eye icon opens an inline content viewer for text-based files (md, txt, csv, json, yaml, etc.)
• Download button on all files
• System docs (README, /docs/*) get view + download but no delete button — they're read-only in the UI
• User uploads get view + download + delete as before

Files changed

• admin/app/controllers/rag_controller.ts
• admin/app/services/rag_service.ts
• admin/app/validators/rag.ts
• admin/start/routes.ts
• admin/types/rag.ts
• admin/inertia/components/chat/KnowledgeBaseModal.tsx
• admin/inertia/lib/api.ts

[chriscrosstalk]

Thanks @bravosierra99 — the feature itself (view + download + metadata columns for KB files) is solid and genuinely useful. Good security posture too (path-traversal guards, restricting viewable types to text, no actions on system-owned docs).

One problem though: your branch is significantly behind current dev, and the PR is pulling forward ~75 files worth of changes that have already shipped in v1.31.0 and v1.31.1. The real feature is probably 300–500 lines across ~7 files, but the diff currently reads as 3,318 additions across 85 files including the PNG-to-WEBP migration (#575), Kiwix library-mode migration (#622), Ollama overhaul (#645/#649/#744), the RAG pipeline fixes (#745), .tmp download staging (#448), the Community Add-Ons docs page (#753), workflow updates, a CONTRIBUTING rewrite — all of which are already on dev.

GitHub is marking it DIRTY/CONFLICTING as a result, and there's no practical way to review this until the noise is separated from the signal.

Could you rebase onto current dev and narrow the PR to just the feature? The files I'd expect to see in a clean version:

• admin/app/controllers/rag_controller.ts (the two new endpoints)
• admin/app/services/rag_service.ts (getStoredFiles() returning StoredFile[])
• admin/app/validators/rag.ts (new validators)
• admin/start/routes.ts (route registration)
• admin/types/rag.ts (StoredFile type)
• admin/inertia/components/chat/KnowledgeBaseModal.tsx (UI)
• admin/inertia/lib/api.ts (client helpers)

Cleanest path is usually to create a new branch off current origin/dev and cherry-pick just your feature commits onto it, then force-push over this branch. If you hit a wall, say the word and I'll walk through it with you.

Appreciate the work — want to get this one landed, just need the diff manageable first.

[bravosierra99]

I rebased again, and pushed some updates... I think my testing wasn't quite on exactly what was in the PR. It should be up to date from now and I did some quick sanity checking of the functionality. There should be no remaining conflicts. Let me know if this works for you @chriscrosstalk

[jakeaturner]

@bravosierra99 great stuff here - just a couple quick changes and this will be ready to merge. Thanks for the contribution!

[jakeaturner]

Existing formatBytes utility available from ~/lib/util -- import from there for DRY

[jakeaturner]

Good baseline for path traversal protection but technically still allows reading from anywhere in the app directory (process.cwd()). We should follow pattern from docs_service or even similar check in rag_service for tighter scoping and consistency

[bravosierra99]

updates made.

[bravosierra99]

▎ @jakeaturner heads up — I force-pushed this branch with a full rebuild on current dev rather than a
▎ fix-and-rebase. RFC #883 reworked the KB ingest tracking around StoredFileInfo + state machine, so
▎ the old commits didn't apply cleanly; rebuilding was simpler than reconciling line-by-line.
▎
▎ Both of your prior comments are addressed in the rebuild:
▎
▎ - formatBytes now imported from ~/lib/util instead of a local copy.
▎ - Path-traversal guard rewritten to match the docs_service pattern — scope is also tighter than
▎ before: only files under RagService.UPLOADS_STORAGE_PATH are viewable/downloadable. ZIM files and
▎ admin docs are read-only in the UI.
▎
▎ Other changes vs. the version you reviewed:
▎ - View/Download buttons gated on isUserUpload && size != null.
▎ - Sortable columns sort within each bucket so the zim/upload/admin_docs grouping from dev is
▎ preserved.
▎ - 7 new unit tests in kb_file_grouping.spec.ts covering the sort + bucket-order invariants.
▎
▎ Built and tested locally against current dev; CI green. Ready for another look when you have a
▎ minute.

[chriscrosstalk]

Thanks @bravosierra99 — the file viewer/download piece here is still very much wanted (it's the core of #952). Heads up though: this branch predates the Knowledge Base ingestion rework that shipped in v1.32.0, and there's significant overlap. On current dev, getStoredFiles() already returns a rich StoredFileInfo[] (with per-file ingest state) instead of string[], and KnowledgeBaseModal.tsx was substantially reworked, so this no longer merges cleanly.

Could you rebase onto current dev? The valuable, non-overlapping part is the new /files/content and /files/download endpoints and the inline viewer — those slot in nicely on top of the new StoredFileInfo[] shape. Happy to review once it's rebased. Thanks for your patience on this one.