settings.toml

[paths]
# Directory to index (relative to project root or absolute)
root_dir = "./examples"
# SQLite database file for passages + vectors
db_path  = "db/index.db"

[model]
name = "all-MiniLM-L6-v2" # Hugging Face model slug loaded by SentenceTransformer
embed_dim  = 384 # Dimension of the embedding vectors (must match the model)

[index]
chunk_len      = 1000   # Passage chunk length
chunk_overlap  = 40    # Overlap
top_k          = 5      # Top‑K search results returned to the user
pool_size      = 20     # candidates from each list, set to 0 to effectively turn off BM25
rrf_k          = 60     # larger K flattens the rank curve
line_by_line_ext = [    # extensions to index line‑by‑line
  ".csv", ".tsv"
]

[filters]
max_file_size_mb = 5    # Skip files larger than this many megabytes
# Explicitly ignore these file extensions (case‑insensitive)
blacklist_ext = [
  ".jpg", ".jpeg", ".png", ".gif", ".bmp", ".tif", ".tiff",  # images
  ".pdf", ".doc", ".docx", ".ppt", ".pptx", ".xls", ".xlsx", # office
  ".zip", ".tar", ".gz", ".7z", ".rar",                      # archives
  ".exe", ".dll", ".so", ".dylib",                           # binaries
]
# Also skip directories with these names (regardless of depth)
blacklist_dirs = ["__pycache__", ".git"]