Release Notes
Fixed
- fastembed cache leak (P0): model weights (
.fastembed_cache/, ~23 MB) were
written into the developer's repo root on every gcx serve. Cache now lives at
$XDG_DATA_HOME/gitcortex/models — fully machine-local, invisible to developers.
Added .fastembed_cache/ to .gitignore and .gitcortex/ignore as a backstop.
- Semantic index version check: format version was silently ignored on load, so
changing the node text representation had no effect. Version mismatches now force
a clean rebuild. Format version bumped to 2.
Changed
- Richer semantic embeddings:
node_text now appends identifier-tokenised words
(CamelCase/snake_case split into lowercase tokens) alongside the qualified name and
signature. NL queries like "validate token" now match validate_token without
relying on the model to unsplit glued identifiers.
- Scaled semantic scoring: semantic hits are scored by actual cosine similarity
mapped to [40‥70] instead of a fixed 45. A cosine-0.95 hit ranks near a prefix
match; a cosine-0.51 hit ranks below token matches — proportional confidence.
- Dedup by node ID: semantic hits were previously deduplicated by symbol name,
silently dropping same-named symbols from different modules. Dedup is now by
qualified name, so all variants surface.
Install gitcortex 0.5.1
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/bharath03-a/GitCortex/releases/download/v0.5.1/gitcortex-installer.sh | sh
Download gitcortex 0.5.1
