Index configuration

What gets indexed, and how, lives in brain/index.config.json under the root. The app’s index settings (i) edit it, and the CLI and MCP server read it, so the index you tune is the index your agents recall against. Without the file, the defaults index the root folder only.

{
  "version": 1,
  "sources": [
    { "id": "workspace", "label": "This workspace", "path": ".", "enabled": true,
      "include": [], "exclude": [], "content": true, "maxDepth": 0 },
    { "id": "notes", "label": "Notes", "path": "~/Documents/Notes", "enabled": true,
      "include": ["*.md", "*.pdf"], "exclude": ["archive/"], "content": true, "maxDepth": 0 }
  ],
  "exclude": [],
  "ignoreDirs": [],
  "extract": true,
  "content": { "enabled": true, "maxTokensPerDoc": 400, "maxBytes": 2097152 },
  "weights": { "topic": 8, "catalogue": 4, "filename": 2, "content": 1 },
  "topics": {},
  "maxFilesPerSource": 50000,
  "refreshSecs": 10
}

Sources

Field	Meaning
id	Short slug. Files from any source but the first are addressed as @<id>/….
path	Folder to walk: relative to the root, absolute, or ~/….
include	When non-empty, a file must match one of these patterns.
exclude	Files matching any of these are skipped.
content	Index contents, not just paths.
maxDepth	Directory levels to descend; 0 is unlimited.

Everything else

Field	Meaning
exclude	Patterns skipped in every source.
ignoreDirs	Directory names skipped anywhere, on top of the built-in list (node_modules, .git, …).
extract	Pull text from PDF, DOCX, RTF, ODT and HTML with system tools.
content.maxTokensPerDoc	Distinct words kept per document, so one huge file can’t dominate.
content.maxBytes	Larger files are indexed by path only.
weights	How much a match counts in a topic, a router description, a file name, or content.
topics	Keyword → path that always answers it.
maxFilesPerSource	Safety cap per source.
refreshSecs	How long an index is reused before re-scanning; 0 means only on demand.