This release adds 5 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+2 more
Affected surfaces
Summary
AI summaryAdded TurboQuant 4‑bit vector index backend and multi document‑root support.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Low |
Added TurboQuant 4‑bit vector index backend (replaces hnsw_rs) Added TurboQuant 4‑bit vector index backend (replaces hnsw_rs) Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Low |
Added multi document‑root support Added multi document‑root support Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Low |
Added heading‑level markdown chunking Added heading‑level markdown chunking Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Low |
Added PDF form‑feed → `--- Page N ---` marker conversion Added PDF form‑feed → `--- Page N ---` marker conversion Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Low |
Added persistent embedding cache for full rebuilds Added persistent embedding cache for full rebuilds Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Low |
Added `alcove path` command and doctor path summary Added `alcove path` command and doctor path summary Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Low |
Added isolated eval corpus with `--corpus` flag for CI/regression testing Added isolated eval corpus with `--corpus` flag for CI/regression testing Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Feature | Low |
Added IR metrics: NDCG@K, MAP@K, MRR alongside existing Precision@K and Recall@K Added IR metrics: NDCG@K, MAP@K, MRR alongside existing Precision@K and Recall@K Source: granite4.1:30b@2026-06-11-audit Confidence: low |
— |
| Feature | Low |
Added chunk-level precision evaluation for heading-based section accuracy Added chunk-level precision evaluation for heading-based section accuracy Source: granite4.1:30b@2026-06-11-audit Confidence: low |
— |
| Feature | Low |
Added regression detection with configurable thresholds (`--baseline`, `--save-baseline`) Added regression detection with configurable thresholds (`--baseline`, `--save-baseline`) Source: granite4.1:30b@2026-06-11-audit Confidence: low |
— |
| Feature | Low |
Added baseline reference benchmark (`benches/baseline.json`) Added baseline reference benchmark (`benches/baseline.json`) Source: granite4.1:30b@2026-06-11-audit Confidence: low |
— |
| Dependency | High |
Replaced openssl-sys with rustls for all remaining dependencies Replaced openssl-sys with rustls for all remaining dependencies Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Dependency | Medium |
Upgraded llm-kernel from 0.3 to 0.3.5 Upgraded llm-kernel from 0.3 to 0.3.5 Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Dependency | Medium |
Migrated to llm-kernel 0.3.5 `vector-index` feature Migrated to llm-kernel 0.3.5 `vector-index` feature Source: llm_adapter@2026-06-11 Confidence: high |
— |
| Bugfix | Medium |
Fixed file-level grep matching and deferred ONNX loading Fixed file-level grep matching and deferred ONNX loading Source: granite4.1:30b@2026-06-11-audit Confidence: low |
— |
| Bugfix | Medium |
Fixed ambiguous env detection and tilde expansion in config paths Fixed ambiguous env detection and tilde expansion in config paths Source: granite4.1:30b@2026-06-11-audit Confidence: low |
— |
| Bugfix | Medium |
Fixed duplicate project-resolution block in MCP handler Fixed duplicate project-resolution block in MCP handler Source: granite4.1:30b@2026-06-11-audit Confidence: low |
— |
Full changelog
Release Notes
Added
- TurboQuant 4-bit vector index backend (replaces hnsw_rs) (#30)
- Multi document-root support (#27)
- Heading-level markdown chunking (#28)
- PDF form-feed →
--- Page N ---marker conversion (#31) - Persistent embedding cache for full rebuilds
alcove pathcommand and doctor path summary (#26)- Isolated eval corpus with
--corpusflag for CI/regression testing - IR metrics: NDCG@K, MAP@K, MRR alongside existing Precision@K and Recall@K
- Chunk-level precision evaluation for heading-based section accuracy
- Regression detection with configurable thresholds (
--baseline,--save-baseline) - Baseline reference benchmark (
benches/baseline.json)
Changed
- Migrated to llm-kernel 0.3.5
vector-indexfeature - Upgraded llm-kernel from 0.3 to 0.3.5
- Replaced openssl-sys with rustls for all remaining dependencies
Fixed
- File-level grep matching and deferred ONNX loading
- Ambiguous env detection and tilde expansion in config paths
- Duplicate project-resolution block in MCP handler
Install alcove 0.12.0
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/epicsagas/alcove/releases/download/v0.12.0/alcove-installer.sh | sh
Install prebuilt binaries via powershell script
powershell -ExecutionPolicy Bypass -c "irm https://github.com/epicsagas/alcove/releases/download/v0.12.0/alcove-installer.ps1 | iex"
Install prebuilt binaries via Homebrew
brew install epicsagas/tap/alcove
Download alcove 0.12.0
| File | Platform | Checksum |
|--------|----------|----------|
| alcove-aarch64-apple-darwin.tar.xz | Apple Silicon macOS | checksum |
| alcove-x86_64-pc-windows-msvc.zip | x64 Windows | checksum |
| alcove-aarch64-unknown-linux-gnu.tar.xz | ARM64 Linux | checksum |
| alcove-x86_64-unknown-linux-gnu.tar.xz | x64 Linux | checksum |
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About epicsagas/alcove
MCP server that gives AI coding agents on-demand access to private project docs via BM25 ranked search. One setup for Claude Code, Cursor, Codex, Gemini CLI, and more. Docs stay private, never in public repos.
Related context
Related tools
Earlier breaking changes
- v0.10.0 Changes code indexing to index all available languages by default instead of auto‑detecting
Beta — feedback welcome: [email protected]