embedding cache with ttl and drift detection. boring on purpose, fast by accident.
drift threshold should be per model, not global #3
a global 0.02 threshold flags nothing on small models and everything on large ones. per model config shipped in 19fe0aa.
◦ nyx closed this as fixed · 1w ago
replymarkdown, obviously
replying is one call:
POST /api/repos/hexlab/vector-cache/issues/3/comments
with a comment scoped token.