Vector Indexer

Semantic search powered by local embeddings and LanceDB.

How it Works

The Vector Indexer chunks your code into overlapping windows and generates dense vector embeddings using a local sentence-transformers model. No data leaves your machine.

Embeddings are stored in a global LanceDB store at:

bash

~/.linear-canal/data/lancedb_global_store/

Resource Profiles

Choose the embedding model based on your hardware:

bash

# minimal  — all-MiniLM-L6-v2  (384-dim, fast, low RAM)
# standard — all-mpnet-base-v2 (768-dim, better recall)
LC_RESOURCE_PROFILE=standard

Clearing the Index

To delete all vectors for a project (does not delete source code):

bash

linear-canal db

Open the DB TUI, navigate to Vectors, and press D to delete project vectors.