OpenBanking Finance v2

Self-hosted personal finance dashboard for Australian bank and brokerage accounts. Syncs via Redbark (open-banking + brokerage), stores everything in a local SQLite database, and renders six tabs of charts — spending, net worth, cash flow, insights, trends, and a financial-independence forecast. One process, no build step, no third party sees your data after the sync hop.

All six tabs, captured against a generated demo dataset (see scripts/seed_demo.py). Dark and light themes are both supported.

Tab	Preview
Spending — by-category, top merchants, scrollable transaction log
Net worth — total + per-account, balance snapshots over time
Cash flow — monthly income vs spend, budget editor
Insights — calendar heatmap, income→category sankey, treemap, day-of-week × hour heatmap
Trends — rolling 30-day spend, monthly category mix, day-of-month
Forecast — net-worth projection and FI calculator

• FastAPI + uvicorn — JSON API + static dashboard files
• SQLite (WAL) via SQLModel
• APScheduler — in-process poller (15-minute interval, in the FastAPI event loop)
• httpx — async Redbark REST client (rate-limit aware, 4-concurrent semaphore for heavy endpoints)
• Apache ECharts (CDN) — bespoke vanilla HTML/JS dashboard, no build step
• pydantic-settings + python-dotenv — config

Pinned to Python 3.12. uv is used if available; otherwise pip + venv.

make install
cp .env.example .env
# edit .env and set REDBARK_API_KEY=...

Get an API key from your Redbark account (Developer or Professional plan required for API access).

# API + scheduler (combined process)
make api

# One-off historical backfill (default: 2 years back)
make backfill ARGS="--from 2024-01-01"

# Backfill a single account, skipping trades
make backfill ARGS="--from 2024-01-01 --account-id <uuid> --no-trades"

The dashboard is at http://localhost:8000/. JSON endpoints live under /api/.

1. make install
2. Set REDBARK_API_KEY in .env
3. Run make api once so the DB is created and the first poll populates connections + accounts. Stop after one poll cycle (15 min) or use python -m app.poller --once.
4. Run make backfill ARGS="--from 2024-01-01" to pull two years of history. After backfill, the per-account watermarks are advanced to "now" so the regular poller will not re-fetch history.
5. Restart make api and let the scheduler take over.

curl -X POST http://localhost:8000/api/budgets \
    -H "Content-Type: application/json" \
    -d '{"category": "FOOD_AND_DRINK", "monthly_limit": "800.00", "currency": "AUD"}'

Or use the Cash flow tab — budget editing is wired into the dashboard.

make check       # lint + typecheck + tests
make fmt         # ruff format + autofix
make test        # pytest only
make typecheck   # mypy --strict
make lint        # ruff check

• Watermarks: each account has last_polled_transactions_at and last_polled_trades_at. Each poll requests from = watermark - 24h so late-posting / re-categorised rows refresh. Idempotent merge on provider ID prevents duplicates.
• Rate limits: heavy endpoints (/transactions, /trades, /holdings) are gated by an asyncio.Semaphore(4). The client self-throttles when X-RateLimit-Remaining drops below 5 and respects Retry-After on 429s.
• Truncation: X-Redbark-Truncated: true is logged but does not stop pagination — the loop keeps requesting the next offset.
• Money: Decimal end-to-end. JSON serialisation emits Decimals as strings.

• Single-user, single-machine. No auth on the dashboard.
• SQLite — fine for one user, not suitable for multi-tenant.

The screenshots above come from a generated dataset, not real bank data. To regenerate:

# Seed a fresh demo.db (~14 months of plausible transactions, balances, trades)
DATABASE_URL=sqlite:///./demo.db PYTHONPATH=. uv run python scripts/seed_demo.py

# Serve the dashboard against demo.db with the poller disabled
PYTHONPATH=. uv run python scripts/serve_demo.py    # → http://127.0.0.1:8765/

# Capture all six tabs in both themes via Playwright + system Chrome
PYTHONPATH=. uv run python scripts/capture_screenshots.py