FirmVault Search

FirmVault Search

Retrieve documents from the FirmVault case corpus by semantic similarity + structured filters. Backed by qmd query and a Python wrapper that adds frontmatter-based filters and case-scope auto-detection.

When to use

• "Find Dr. Smith's notes in the Whaley v. Acme case"
• "Show me all medical records from October 2025"
• "What did the insurance adjuster say last month?"
• "Search across cases for surgery records mentioning rotator cuff"
• "Find all correspondence with State Farm in the active case"
• "Pull up legal filings from the Acme matter"

How to call

uv run firmvault_search "<query text>" \
  [--scope all|case|<slug>] \
  [--category medical|legal|insurance|communication] \
  [--case <slug>] \
  [--date 2025-10-15] \
  [--date-range 2025-10-01..2025-10-31] \
  [--provider <name>] \
  [--limit 10]

Output is a JSON array — one object per hit:

{
  "file_path": "/Users/.../FirmVault/cases/<slug>/medical/2025-10-15-dr-smith-visit.md",
  "score": 0.87,
  "snippet": "Patient presented with...",
  "frontmatter_excerpt": {
    "case_slug": "whaley-v-acme",
    "document_category": "medical",
    "document_date": "2025-10-15",
    "provider": "Dr. Smith Clinic",
    "document_type": "office_visit_note"
  }
}

Scope auto-detection

If --scope is not given OR --scope=case:

• The CLI walks up from cwd looking for FirmVault/cases/<slug>/. If found, the search scopes to that slug automatically.
• If no such ancestor exists, the search falls back to cross-case (all cases).

Pass --scope=all to force cross-case. Pass --scope=<slug> to target a specific case explicitly.

Filters

• --category — exact document_category match (medical | legal | insurance | communication)
• --case — exact case_slug match (alternative to --scope=<slug>)
• --date — exact document_date match (YYYY-MM-DD)
• --date-range START..END — inclusive document_date range
• --provider — case-insensitive substring match on provider / sender / recipient
• --limit — max hits returned (default 10)

Filters apply Python-side post-filter on the markdown frontmatter, so unindexed fields still work.

Tips

• Read the top hit's file_path with the Read tool to surface details to the user.
• For "find all X" queries, set --limit 25 or higher and summarize.
• Use the snippet for at-a-glance scanning — only Read full files when needed.
• When inside a case directory, omit --scope to get case-scoped results automatically.