~/.clawdex/contacts/, one people/<slug>/person.md per person: YAML frontmatter (name, emails, phones, tags, account handles) plus a freeform markdown bio. This is a direct file source: DuckBrain reads those person.md files directly (no SQLite), one bronze row per person whose text is the whole contact card. A person is searchable by name, any contact point, their tags, or your notes about them.
What gets ingested
- One bronze row per person — the text is the entire contact card, so name, contact points, tags, and your bio notes are all searchable.
- Upsert on a stable per-person id — editing a
person.mdupdates that row in place. - Manual refresh. No auto-refresh lane; run
clawdex import/clawdex person add, thenduckbrain sync contacts.
Facet mapping
Rows are
classification: confidential and never leave the machine.