Repository navigation
Releases
Releases · sadigaxund/codexlr8
CodeXLR8 v0.0.3
Search quality infrastructure: eval framework, fuzzy fallback, embeddings layer, fine-tuning.
Added
evalcommand: measure search quality with Precision@1, MRR, Recall@5--explain/-e: per-token hit counts, zero-match detection, quality signal--grouped/-g: directory-clustered results with--scopehints--scope/-s: SQL LIKE pre-filter for path-prefix search- Fuzzy fallback: difflib Levenshtein correction on zero results (config:
fuzzy: true) - Embeddings layer: opt-in hybrid BM25 + cosine rerank via
embeddings.enabled traincommand: TSDAE fine-tune any sentence-transformers model on codebaserecommend-modelcommand: suggest best model by codebase size- Scope-granularity eval:
file,scope,exactassert modes with line overlap - Incremental embed:
--incrementalonly re-embeds changed files
Changed
- Query semantics: AND-then-OR → pure OR + token-coverage scoring
- Default excludes: added
examples/*,docs/*,tutorials/*,benchmarks/* - JSON output:
{"results": [...]}wrapper with optionalexplain/groupedkeys
Fixed
- Flat BM25 when metadata absent — path weighting provides differentiation
- No path differentiation in ranking — filename, dir component weighting
- AND-then-OR precision hole — multi-token queries no longer fail or return noise
CodeXLR8 v0.0.2
Search quality improvements from benchmark feedback.
Added
--explain/-e: query token diagnostics showing per-token hit counts, filtered words, and quality signal--grouped/-g: clusters search results by directory with--scopehints--group-depth <n>: control directory depth for grouping (default 3)--scope/-s: restrict search to a path prefix, applied as pre-score SQL filter- Path weighting: filename match (0.8×), filename component (0.7×), directory (0.5×) — works without metadata
- Coverage warning:
statuswarns when <10% of files have metadata - Matched tokens: each result includes which query tokens matched
Changed
- Query semantics: replaced AND-then-OR fallback with pure OR + token-coverage scoring
- Default excludes now include
examples/*,docs/*,tutorials/*,benchmarks/* - JSON output format:
{"results": [...]}instead of bare list (supportsexplain,groupedwrappers)
Fixed
- Flat BM25 when metadata absent — path weighting now provides differentiation without sidecars
- No path-weighting in ranking —
lib/foo.pynow outranksexamples/foo.pyfor matching query - AND-then-OR precision hole — multi-token queries no longer return zero or too many flat-score results
CodeXLR8 v0.0.1
Initial release. A purpose-built codebase search engine for LLM coding agents.
Features
- Full-text codebase search with SQLite FTS5 and AND semantics
.meta.yamlsidecar files for curated metadata (summary, tags, public_api)- CLI:
scan,init,index,search,status,setup - MCP server for agent integration (
codebase_search,codebase_index) - Agent skill file with session workflow guidance
- Custom ranking: metadata boosts (public_api > tags > summary) + BM25
- Configurable scope: root, include/exclude patterns, file extensions, ignore dirs
- Incremental indexing via file mtime tracking
- Interactive
setupwizard for.codexlr8.yaml - JSON output for programmatic use