Embedding index for semantic search #14

New issue

Open

opened 2026-04-07 20:23:21 +00:00 by austin · 0 comments

austin commented

2026-04-07 20:23:21 +00:00

Owner

Implement semantic code search using fastembed.

Embed code chunks (functions, classes) and their docstrings/comments at index time
Embed the user's task description at query time
Retrieve the top-K most semantically relevant code chunks for a given query
Complement the AST-based symbol lookup: AST finds by name/structure, embeddings find by meaning
Store embeddings on disk alongside the AST index (under .localcode/)
Incremental updates: re-embed only changed symbols

Use nomic-embed-text or equivalent small embedding model via fastembed. This runs on CPU and should be fast enough for interactive use.

Implement semantic code search using fastembed. - Embed code chunks (functions, classes) and their docstrings/comments at index time - Embed the user's task description at query time - Retrieve the top-K most semantically relevant code chunks for a given query - Complement the AST-based symbol lookup: AST finds by name/structure, embeddings find by meaning - Store embeddings on disk alongside the AST index (under `.localcode/`) - Incremental updates: re-embed only changed symbols Use `nomic-embed-text` or equivalent small embedding model via fastembed. This runs on CPU and should be fast enough for interactive use.