DocVault

Self-hosted personal finance and document workspace. One container holds your tax records, tracks net worth across brokers / crypto / metals / real estate, parses financial PDFs with Claude, ingests Apple Health data from an iOS Shortcut, and surfaces macro + crypto quant signals alongside AI-generated strategy notes — all without sending data off your machine.

Screenshots in this README are captured against the included demo-data/ fixtures — the numbers, entities, and Strategy entry are fabricated.

Try It Locally (Demo Mode)

Want to poke around before connecting real data? The repo ships with a demo-data/ directory and a second Vite config that points the dev server at it.

bun install
# Terminal 1 — demo backend on port 3006, reading demo-data/
DOCVAULT_DATA_DIR=./demo-data \
  DOCVAULT_PORT=3006 \
  DOCVAULT_PASSWORD=demo \
  DOCVAULT_MASTER_KEY=$(openssl rand -base64 32) \
  bun run server/index.ts
# Terminal 2 — demo frontend on port 5174, proxying /api to :3006
vp dev --config vite.demo.config.ts

Open http://localhost:5174 and sign in with admin / demo — full app, fake data. Your ./data/ stays untouched.

Features

Documents & Taxes

• Multi-entity organization — separate spaces for personal, LLCs, property, military, etc.
• AI document parsing — Claude Vision extracts structured data from W-2s, 1099s, K-1s, receipts, bank statements (~$0.003/page).
• Auto file naming — uploads are renamed to {Source}_{Type}_{Date}.ext.
• Type-specific parsers — 15 document-type parsers (W-2, 1099-NEC, K-1, statement, receipt, 1098, Koinly, Schedule C, etc.) normalize results into a single analytics module.
• Federal tax summary — Schedule C, K-1, capital gains, withholdings aggregated across all entities per year.
• Solo 401(k) calculator (IRS Pub 560 worksheet), estimated quarterly tracker, TN state view, mileage log, sales ledger, invoice tracking.
• CPA package export — one click to bundle an entity/year into a ZIP for your accountant.

Net Worth & Portfolio

• Unified portfolio view across every account — brokerage, crypto, banks, metals, real estate.
• Automatic daily snapshots with a historical net worth chart.
• Broker aggregation via SnapTrade (Fidelity, Vanguard, Robinhood, etc.) with per-account history.
• Crypto exchange balances (Kraken, Coinbase, Gemini) + Etherscan wallet scanning across mainnet, Arbitrum, Optimism, Polygon, and Avalanche.
• Precious metals with live spot prices.
• Real estate with cost basis, equity, and property-level notes.
• Bank balances + transactions via SimpleFIN Bridge (16,000+ US institutions).

Quant Dashboards & Strategy

• Quant section with 28 endpoints powering dashboards for crypto (BTC risk, hash ribbons, drawdown), macro (Sahm rule, yield curve, NFCI, fed stance, recession probability), housing, GDP & growth, commodities, VIX term structure, global markets, and an auto-generated macro event calendar.
• Strategy history — AI-generated investment strategy notes authored by Claude Code via a /strategy skill that reads your portfolio + current quant signals and saves a regime-aware recommendation. Entries render as expandable cards with a signal grid and full markdown analysis (tables, allocations, action plans). Dollar amounts and percentages in the markdown are obscured when "Blur financial numbers" is on.

Politics & Congressional Trading

In-house ingest of congressional and executive-branch disclosures — no external service, and no API key for the trading data (it's all public government filings).

• Politician stock trades — House & Senate Periodic Transaction Reports plus Trump's OGE-278-T disclosures, parsed into one normalized feed. Scanned / hand-filed PTRs that pdftotext can't read are recovered by checkbox-form OCR (gridline detection + per-cell pixel reading; poppler + tesseract are baked into the Docker image).
• Options contract detail — strike, expiry, call/put, and contract count are pulled from the filing's free-text DESCRIPTION field (e.g. Pelosi's "Purchased 20 call options, strike $150, exp 1/15/27"), not just the underlying ticker.
• Consensus clustering — surfaces when several members buy (or sell) the same ticker in the same direction within a window.
• Copy-trade backtest — a performance leaderboard: "if you'd mirrored each politician's stock buys at the disclosed size, where would you be now?" Stock buys get real P&L (exact share counts when the filer states them, else estimated from the amount range and flagged); options report the underlying's move (the contract isn't priced). Recomputed daily; prices via yahoo-finance2 (keyless).
• Filings archive — every fetched PDF + extracted text + metadata is saved under the data dir, searchable and re-parseable without re-fetching.
• Full-screen browse — click a dashboard metric to search/filter/see-all of trades, bills, executive actions, or archived filings.
• Bills & executive actions — recent Congress.gov bills (the one piece needing a free Congress.gov API key, set in Settings) + presidential executive actions (keyless, via the Federal Register).
• Self-hosted member headshots — portraits downloaded once and served from DocVault, not hot-linked.

Populate it: the feed refreshes daily on a forward-only schedule (new filings only). To pull the current year's history in one pass, POST /api/politics/backfill — it runs server-side (poll /api/politics/feed for progress). Everything except the Bills stream works with zero credentials.

⚠️ Disclosures carry a ~45-day legal reporting lag and report dollar ranges, not exact sizes — the backtest is honest about both (estimates flagged, options labeled as the underlying's move).

AI Chat

Heavily inspired by t3.chat — a multi-thread Claude chat that can read across your entire vault. The sidebar lists every thread; the active conversation streams in the main panel with markdown rendering, image/PDF attachments, and tool calls shown as collapsible cards.

• Claude OAuth subscription token — paste the token from claude setup-token and chats are billed to your Claude.ai subscription instead of the API. Falls back to an API key if you'd rather pay per-token.
• Voice input via Parakeet (or any OpenAI-compatible transcription service) — point Settings → Chat & Voice at a /audio/transcriptions endpoint such as parakeet-mlx, faster-whisper-server, or lightning-whisper-mlx running on your LAN. Push-to-talk in the composer; audio never leaves your network.
• Multi-thread sidebar — threads persist locally; switch, rename-by-derivation, delete, or start fresh without losing context.
• Tool-using agent — Claude can list entities, read files, search documents, compute tax summaries, and tag/note files. Each tool call renders as an expandable card in the response.

Health (Apple Health)

• iOS Shortcut daily sync — a one-tap shortcut pushes HealthKit data to DocVault's /api/health/ingest endpoint.
• Multi-person support — each member of the household gets their own snapshot.
• Overview + per-segment dashboards — activity, heart, sleep, workouts, body composition.
• Automatic illness detection — rolling baseline over wrist temperature, heart rate, and HRV flags probable illness windows.
• Running ROI (vs BTC/SPX), workout segment insights, sleep quality scoring, recovery scoring.

Time Tracking & Billing

• Timesheet — customers → projects → time entries, with per-project rates snapshotted onto each entry so later rate changes never rewrite billing history.
• Two ways to log time — a start/end span, or just hours ("2.5h doing X") when you didn't watch the clock. Both produce the same entry; only the span-based one appears on the hour-of-day punchcard.
• Sub-clients — optional sub-divisions of a project (an end client, matter, workstream, or cost code). Tag an entry with one and every report groups by it. Archive a sub-client to retire it without disturbing the entries that already reference it.
• Invoices — generated from uninvoiced entries, rendered to PDF from a reusable template (sender identity, payment terms, bank details), with per-project retainer floors that add a labeled top-up line. Invoice PDFs carry the labels bill-pay systems parse (Invoice #:, Due Date:, a page-one Amount Due:), so multi-page invoices can't be mis-read by page-limited OCR.
• Send invoices by email with the PDF attached, from an editable draft you always review first.
• Scheduled timesheet reports — email a categorized summary of recent entries to a billing contact on a schedule, as HTML plus a CSV attachment. See below.

Scheduled timesheet reports

Timesheet → Weekly Report configures a recurring email so whoever pays you can drop your hours straight into their own billing run.

Setting	What it does
Cadence	Every week, every 2 weeks, or every month
Day + hour	When in that period it sends (plus an optional timezone override)
Window	How many days back the report covers — defaults to 7
Recipients	Comma-separated; the same report goes to all of them
Scope	Which clients (and optionally which projects) the recipient may see
Categories	Keyword rules that bucket entries when they aren't tagged

Scope is the important one. A report goes to one client's billing contact, so leaving scope empty means that recipient sees every client's hours. Pick the clients this recipient is entitled to.

Each entry's category is resolved most-trustworthy-source-first:

1. its sub-client, if tagged — a recorded fact
2. the first keyword rule that matches its description or project — an inference
3. the project name — the honest fallback

Prefer sub-clients over keyword rules. A long entry covering several tickets matches whichever rule is listed first, which is how a week of infrastructure work ends up filed under "Meetings."

The report is read-only over your entries: it never marks anything invoiced and never creates an invoice, so it's independent of the invoicing flow. It includes non-billable entries (flagged as such) and doesn't care whether an entry has already been billed. Preview any window before enabling anything, download the CSV, or send off-cycle with Send now — a manual send records the same watermark, so it suppresses that period's automatic delivery instead of double-sending.

Backup, Sync, Observability

• Encrypted backup/restore — AES-256-GCM zip of all config + parsed data, downloadable on demand.
• Scheduled auto-backup runs before each Dropbox sync.
• Dropbox sync — rclone-based one-way push of every entity folder on a configurable schedule (default 15 min).
• Custom sync paths — drop .docvault-dropbox-map.json to map entities to specific Dropbox folders.
• Portfolio snapshot scheduler + Dropbox sync scheduler configurable from Settings.
• System Status panel (Settings → System Status) — scheduler timers, next-run times, last error, and a live log viewer.

Privacy

• Blur financial numbers toggle (Settings → Preferences) — obscures every dollar amount and percentage across the UI, including AI-generated markdown in the Strategy section.
• Your data never leaves your machine — no telemetry, no analytics, no remote database. One volume mount, one container.

Other

• Mileage log with address autocomplete.
• Calendar: birthdays ("turns N"), one-offs, and recurring tasks with fixed or after-completion scheduling — month grid, agenda, chat tools, and a "Week Ahead" box in the Daily News edition. Absorbs the old deadline reminders (existing data migrates automatically).
• Username/password auth with session cookies.
• Docker-ready: single container, auto-published to GHCR (amd64 + arm64).

Quick Start

bun install
bun start

Frontend: http://localhost:5173 — Backend: http://localhost:3005

Storage Setup

mkdir -p data/personal data/my-llc data/property
# or symlink existing folders
ln -s ~/Documents/taxes data/personal

Docker

docker run -p 3005:3005 \
  -v /path/to/documents:/data \
  -e ANTHROPIC_API_KEY=sk-ant-... \
  -e DOCVAULT_PASSWORD=yourpassword \
  ghcr.io/vanities/docvault:latest

Docker Compose

services:
  docvault:
    image: ghcr.io/vanities/docvault:latest
    ports:
      - '3005:3005'
    volumes:
      - /path/to/documents:/data
    environment:
      - ANTHROPIC_API_KEY= # Required for AI parsing
      - DOCVAULT_USERNAME=admin # Default: admin
      - DOCVAULT_PASSWORD= # Required
    restart: unless-stopped

Environment Variables

Variable	Required	Description
ANTHROPIC_API_KEY	For AI parsing	Claude Vision API key
DOCVAULT_USERNAME	No	Login username (default: admin)
DOCVAULT_PASSWORD	Yes	Login password; server startup fails closed without this unless opted out
DOCVAULT_ALLOW_UNAUTHENTICATED	No	Explicit local/demo-only opt-out (true, 1, or yes)
DOCVAULT_DATA_DIR	No	Data directory path (default: ./data)
DOCVAULT_PORT	No	Backend port (default: 3005)

All integrations (SimpleFIN, SnapTrade, Etherscan, Kraken, Coinbase, Gemini, Dropbox) are configured through Settings and stored in data/.docvault-settings.json.

Data Files

Everything lives in DOCVAULT_DATA_DIR as .docvault-*.json — mount one volume and the install is portable:

File	Purpose
.docvault-config.json	Entity definitions
.docvault-settings.json	API keys and integration config
.docvault-parsed.json	Cached AI parse results
.docvault-metadata.json	Document tags and notes
.docvault-calendar.json	Calendar events and completions
.docvault-portfolio-snapshots-*.json	Yearly portfolio snapshot history
.docvault-broker-cache.json	Cached brokerage aggregation
.docvault-crypto-cache.json	Cached crypto balances
.docvault-simplefin-cache.json	Cached bank balances
.docvault-gold.json	Precious metals holdings
.docvault-property.json	Real estate portfolio
.docvault-strategy-history.json	Saved Strategy entries
.docvault-health.json	Apple Health ingested data
.docvault-sync-status.json	Dropbox sync status

Tech Stack

Layer	Technology
Frontend	React 19 + TypeScript + Tailwind CSS (Vite+)
Backend	Bun native server (Bun.serve())
Storage	Local filesystem
AI	Anthropic Claude Vision API
Health	iOS Shortcuts + HealthKit ingest
CI/CD	GitHub Actions → GHCR

License

GNU General Public License v3.0