Indexes a Go project with Tree-sitter, creates Jina code embeddings with Candle, and stores functions, methods, structs, and interfaces in SQLite using SQLite-Vector.

On the first run, slopmop automatically downloads SQLite-Vector 1.1.0 for the current platform and stores it in the system cache. To use an existing or custom build instead, set SQLITE_VECTOR_PATH to its shared library:

export SQLITE_VECTOR_PATH=/absolute/path/to/vector.dylib # macOS
# export SQLITE_VECTOR_PATH=/absolute/path/to/vector.so  # Linux
# set SQLITE_VECTOR_PATH=C:\absolute\path\to\vector.dll # Windows

Build or install the executable:

cargo install --path .

Then run it from a Go project:

cd path/to/go-project
slopmop

Alternatively, provide the project directory while developing slopmop:

cargo run --release -- path/to/go-project

The index is created as .slopmop in the project root. Each run recursively finds .go files and atomically replaces the existing index. The first run downloads and caches jinaai/jina-embeddings-v2-base-code from Hugging Face.

Add a .slopmopignore file to the project root to exclude paths from indexing. It uses gitignore syntax, including comments, glob patterns, directory patterns, root-relative patterns, and ! negation:

# Generated code
internal/generated/
*.generated.go

# Ignore one file only at the project root
/legacy.go

# Re-include an otherwise ignored file
!important.generated.go

Ignored directories are not traversed. As with .gitignore, a file inside an ignored directory cannot be re-included unless its parent directory is also re-included.

From an indexed project, list the 10 largest similarity clusters:

slopmop cluster

The default minimum cosine similarity is 0.8. It can be changed explicitly, and a project directory can be supplied:

slopmop cluster path/to/go-project --threshold 0.75

Nodes are connected when their cosine similarity meets the threshold. Transitive connected nodes belong to the same cluster.

The database contains two tables:

files
  id        INTEGER PRIMARY KEY
  filepath  TEXT NOT NULL UNIQUE

embeddings
  id         INTEGER PRIMARY KEY
  embedding  BLOB NOT NULL
  node_name  TEXT NOT NULL
  file_id    INTEGER NOT NULL REFERENCES files(id)

Embeddings are normalized 768-dimensional FLOAT32 vectors. The vector column uses cosine distance and can be queried with SQLite-Vector's vector_full_scan or, after quantization, vector_quantize_scan.