config.example.toml

# NoteBrain CLI Configuration Template
#
# Copy this file to:
#   - ~/.notebrain/config/config.toml (Default global config location)
#   - Or any path and reference it using: notebrain --config /path/to/config.toml
#
# Note: Command line flags always override values specified in this file.

# ------------------------------------------------------------------------------
# Global Settings
# ------------------------------------------------------------------------------

# Path to the local directory where ChromaDB stores its data.
# Defaults to "~/.notebrain/chroma"
chroma-path = "~/.notebrain/chroma"

# Absolute path to your Obsidian vault on disk.
# If set here, you don't need to specify --vault-path on every 'ingest' execution.
# vault-path = "/home/user/Documents/Obsidian Vault"

# Enable verbose debug logs to stderr.
# Defaults to false
verbose = false

# Disable OSC 8 terminal hyperlinks in output.
# Useful if your terminal emulator does not support clickable link rendering.
# Defaults to false
no-hyperlinks = false

# Default CLI output formatting option.
# Options: "text" (human-readable), "json" (structured API), "tsv" (tabular), "ndjson" (newline-delimited JSON)
# Defaults to "text"
format = "text"

# Include full matched note chunk text in structured JSON, TSV, or NDJSON outputs.
# Defaults to false
include-text = false

# Suppress similarity search results below this similarity threshold.
# Must be a floating-point value between 0.0 and 1.0.
# Defaults to 0.0
min-score = 0.0

# Respect Obsidian's userIgnoreFilters and attachmentFolderPath configuration.
# If true, files matching these rules or located in the attachment directory are skipped during ingest.
# Defaults to true
respect-exclude = true

# Use your preferred terminal/GUI editor ($EDITOR environment variable) as the default opener.
# If true, pressing Enter in the interactive search UI opens notes in the editor instead of Obsidian.
# Defaults to false
use-editor = false

# ------------------------------------------------------------------------------
# Ingestion Pipeline Settings
# ------------------------------------------------------------------------------

# Number of concurrent workers to use when reading and parsing markdown files.
# Defaults to 4
workers = 4

# Maximum number of runes (Unicode characters) per chunk sent to the parser.
# MiniLM-L6-v2 is optimal at 128-256 tokens. At ~4.5 chars/rune for English prose,
# 800 runes ≈ 178 tokens — leaving ~78 tokens of headroom for the contextual
# augmentation prefix (title + heading path) prepended before embedding.
# Defaults to 800
chunk-size = 800

# Number of runes repeated at the start of each sub-chunk when a section is split.
# Provides 1-2 sentences of shared context at sub-chunk boundaries, reducing
# information loss when a single large section exceeds chunk-size.
# Defaults to 100
chunk-overlap = 100

# Skip text chunks containing fewer than this many words during ingestion.
# Filters out heading-only sections (e.g. "## Installation"), code-only placeholders,
# and wikilink-only paragraphs that produce near-zero-signal embeddings.
# Rule of thumb: 1 word ≈ 1.3 tokens; MinChunkWords=10 ≈ 13 tokens minimum.
# Defaults to 10
min-chunk-words = 10

# Skip image and file attachment links (e.g., ![[image.png]] or [[file.pdf]]) during indexing.
# If true, attachment links are ignored and excluded from the wikilink graph.
# Defaults to true
skip-attachments = true

# ------------------------------------------------------------------------------
# Search and Traversal Settings
# ------------------------------------------------------------------------------

# Default maximum number of results to return for search, boosted, and hidden commands.
# Defaults to 10
limit = 10

# Maximum number of chunks to return per note in semantic search.
# Can be lowered (e.g., to 1 or 2) to increase result diversity across different notes.
# Defaults to 3
top-k = 3

# Skip uncreated "phantom" notes (notes linked via wikilinks that do not exist as markdown files on disk).
# If true, phantom notes are excluded from graph results (backlinks, connections, tags).
# Defaults to true
skip-phantom = true