Repository navigation
Releases
v0.4.0 - Fast Keyword Search Default
Performance Fix
Default Search Mode Changed to Keyword
Due to qmd's architecture limitations, Semantic search is inherently slow (~20s per query).
Benchmark Results:
| Mode | Speed | Why |
|---|---|---|
| Keyword | 0.25s | No AI models, pure BM25 |
| Semantic | ~20s | Loads 2GB models from disk each query |
| Hybrid | ~30s+ | Loads 3 models, may crash |
Why This Happens
Unlike Omnisearch which keeps models in memory, qmd runs as a CLI tool that loads models fresh for each search.
Recommendation
- Use Keyword for fast, exact text matching
- Use Semantic only when you need conceptual search (and can wait)
- Avoid Hybrid (unstable on some systems)
v0.3.9 - Fix RAG Chat Crash
Bug Fixes
RAG Chat Metal GPU Crash Fixed
- Changed RAG Chat search from
query(Hybrid) tovsearch(Semantic) - Fixes fatal error in ggml-metal-context.m:235 during query expansion
- Significantly faster response time
Root Cause
The query command uses 3 AI models (embedding + generation + reranking) which caused Metal GPU crashes on some Macs. The vsearch command only uses embeddings, avoiding the crash.
Performance Note
- Keyword search: Instant (no AI)
- Semantic search: Fast (embeddings only)
- Hybrid search: Slow (may crash on some systems)
If you still experience crashes, try using Keyword or Semantic mode instead of Hybrid.
v0.3.8 - Improved Search UX
What's Changed
Search Improvements
- Default to Semantic Search: Now uses AI-powered semantic search by default instead of keyword matching
- More Results: Increased default limit from 10 to 50 results
- Relevance Filter: Only shows results with 50%+ relevance score
Bug Fixes
- Fixed document opening: Now uses exact file path to open documents instead of link resolution
- Fixes issue where clicking search results would open wrong/empty files
- Especially important when multiple files have similar names
Upgrade Notes
Existing users may need to adjust settings manually or reinstall the plugin to get new defaults.
v0.3.7 - Fix file opening + keyword search
Bug Fixes
1. File Opening Fixed
- Stripped
qmd://CollectionName/prefix from file paths - Clicking search results now opens the correct note in Obsidian
2. Keyword Search Fixed
- Changed default minScore from 0.3 to 0
- BM25 keyword search scores (~0.18) were being filtered out by 0.3 threshold
- All search modes now work correctly
Score Ranges Reference
| Mode | Typical Scores |
|---|---|
| Keyword (BM25) | 0.1 - 0.3 |
| Semantic | 0.5 - 0.9 |
v0.3.6 - Fix empty results handling
Bug Fix
- Fixed 'No results' parsing error: qmd returns plain text
No results found.instead of JSON when search has no matches - Now gracefully handles empty results
This should fix the Failed to parse qmd response error.
v0.3.5 - Fix search timeout
Changes
- Increased timeout: 30 seconds → 5 minutes
- Hybrid mode (qmd query) can take 2+ minutes on first run while models load
- Default mode changed: Hybrid → Keyword
- Keyword search is fast and reliable
- Users can manually switch to Hybrid when needed
Performance Note
| Mode | Speed | First Run |
|---|---|---|
| Keyword | ~1 second | Fast |
| Semantic | ~2 seconds | Fast |
| Hybrid | 1-3 minutes | Slow (model loading) |
Hybrid mode becomes fast after models are loaded into memory.
v0.3.4 - Fix error display
Bug Fixes
- Fixed [object Object] error display: Error messages now properly shown in UI
- Improved error handling with descriptive messages
Upgrade
BRAT: Settings → BRAT → Check for updates
v0.3.3 - Fix qmd status connection check
Bug Fix
- Fixed 'Not Connected' status: The plugin now correctly parses
qmd statusoutputqmd statuscommand doesn't support--jsonflag- Added text parser to extract index info (documents, embeddings, collections)
- Connection status now properly shows 'Connected' when qmd is working
Upgrade
BRAT users: Check for updates in Settings → BRAT
Manual users: Download and replace files in .obsidian/plugins/qmd-ragidian/
v0.3.2 - Fix Search
Bug Fix
- Fixed: Search returning empty results
- Root cause: qmd returns JSON array directly, not object with 'results' property
v0.3.1
Fixes
- Fix One-Click Setup JSON parsing error (qmd collection list doesn't support --json)
Improvements
- API key inputs now use password type for security
- Updated README with better BRAT installation guide