Finance App

Personal finance tracker built on Next.js 16 (App Router), PostgreSQL, and Prisma. Bank statements are ingested automatically from Paperless-NGX via an N8N workflow that uses Gemini to extract structured data from PDF statements.

Stack

• Frontend: Next.js 16 App Router, TypeScript, Tailwind CSS, Recharts
• Backend: Next.js API routes, raw PostgreSQL via pg + @prisma/adapter-pg
• Database: PostgreSQL (postgres-personal container)
• Auth: X-Forwarded-User header (email) set by Traefik forward-auth → mapped to participants.email
• Ingestion: N8N workflow → Gemini 2.5 Flash (PDF parsing) → PostgreSQL

---

Data Model

statements

The top-level document, one row per billing period per account.

Column	Type	Description
id	int	Primary key
bank_name	text	Normalised bank name (e.g. "American Express")
card_name	text	Product name (e.g. "Rewards Travel Adventures")
account_number	text	Account/card number (spaces stripped)
account_type	text	Raw account type string from statement
statement_type	text	Normalised type: Credit Card, Business Card, multi-currency account, etc.
account_holder_name	text	Name on the account if extracted
billing_start_date	date	Period start
billing_end_date	date	Period end — used as the deduplication anchor
opening_balance	numeric	Balance at start of period
closing_balance	numeric	Balance at end of period
total_credits	numeric	Sum of all credits in period
total_debits	numeric	Sum of all debits in period
total_amount_due	numeric	Amount due (credit cards)
minimum_amount_due	numeric	Minimum payment due (credit cards)
payment_due_date	date	Payment due date (credit cards)
credit_limit	numeric	Credit limit (credit cards)
available_credit	numeric	Available credit at statement date
interest_charged	numeric	Interest charged this period (from statement summary)
fees_charged	numeric	Fees charged this period (from statement summary)
currency	text	Statement currency (e.g. AUD, USD)
exchange_rate_to_aud	numeric	FX rate at ingestion time (live from open.er-api.com)
owner_id	int FK → participants	Which person owns this statement
paperless_doc_id	int	Paperless-NGX document ID — deduplication key
tier_used	text	AI model used for extraction (e.g. gemini-2.5-flash)
event_created	bool	Whether a Google Calendar reminder was created for payment due date

Deduplication: unique index on (bank_name, account_number, billing_end_date) prevents re-ingestion of the same period. paperless_doc_id has a separate unique index for Paperless-linked documents.

Credit card detection: statement_type ILIKE '%card%'

---

transactions

One row per line item within a statement. Cascade-deleted when the parent statement is deleted.

Column	Type	Description
id	int	Primary key
statement_id	int FK → statements (nullable)	Parent statement; NULL for manually-entered transactions
owner_id	int FK → participants (nullable)	Owner for manual transactions (no statement); statement-linked transactions derive owner from statements.owner_id
transaction_date	date	Date of transaction
description	text	Raw description from the statement
amount	numeric	Original amount in statement currency
amount_aud	numeric	AUD-converted amount (= amount if already AUD)
transaction_type	text	debit, credit, payment, refund, fee, interest, transfer
merchant_name	text	Raw merchant name extracted by Gemini
merchant_normalized	text	Cleaned/normalised merchant name (Gemini)
location	text	Location if present on statement
foreign_currency_amount	numeric	Original foreign amount if this was an FX transaction
foreign_currency_code	text	Foreign currency code (e.g. USD)
category	text	AI-assigned category (see category taxonomy below)
row_index	int	Position in statement — used for deduplication
reconciled_with_id	int FK → transactions (nullable)	Links a manually-entered transaction to its matching statement transaction after reconciliation
created_at	timestamptz	When the row was inserted — the "import date". For reconciled transactions the UI shows the original manual/CSV created_at, not the statement's

Deduplication: unique index on (statement_id, transaction_date, description, amount, row_index).

Analytics: all spend queries use amount_aud for cross-currency consistency. Split-adjusted queries apply amount_aud * share_percent / 100 where a split exists for the current user.

---

transaction_overrides

User corrections to AI-extracted data. Stored separately to preserve the original extraction.

Column	Type	Description
transaction_id	int FK → transactions (unique)	One override per transaction
merchant_normalized	text	User-corrected merchant name
category_override	text	User-corrected category
notes	text	Free-text notes

All analytics queries use COALESCE(o.category_override, t.category) and COALESCE(o.merchant_normalized, t.merchant_normalized, t.merchant_name) to prefer overrides over AI values.

---

transaction_splits

Shared expense tracking — records that a transaction was split between participants.

Column	Type	Description
transaction_id	int FK → transactions	The transaction being split
participant_id	int FK → participants	Who shares in this transaction
share_percent	numeric(5,2)	Their percentage (1–100)
settled	bool	Whether this share has been settled
settled_at	timestamptz	When it was settled

A transaction can be split across multiple participants. The statement owner's own share is implicit (100 - SUM(other shares)). Analytics queries LEFT JOIN transaction_splits on participant_id = current_user.id — if no split row exists, the full amount belongs to the owner.

---

transaction_tags

Many-to-many join between transactions and tags.

Column	Type
transaction_id	int FK → transactions
tag_id	int FK → tags

---

tags

User-defined coloured labels for ad-hoc transaction grouping beyond the fixed category taxonomy.

Column	Type	Description
id	int	Primary key
name	text (unique)	Tag name
color	text	Hex colour (default #6366f1)

---

participants

People who own statements or share expenses.

Column	Type	Description
id	int	Primary key
name	text (unique)	Display name
email	text (unique)	Login identity — matched against X-Forwarded-User header

---

account_owner_mappings

Persists (bank, account_number) → owner assignments so future ingestion auto-assigns the correct owner without manual intervention.

Column	Type	Description
bank_name	text
account_number	text
owner_id	int FK → participants

Written when a user reassigns a statement owner in the UI. Consulted by the N8N workflow on every new statement insert.

---

rules

Saved auto-categorisation rules. Applied in bulk via the Rules page.

Column	Type	Description
owner_id	int FK → participants	Rule belongs to this user
name	text	Rule label
conditions	jsonb	Array of {field, operator, value} — AND logic
actions	jsonb	{set_category, add_tag_ids, set_merchant}
enabled	bool
priority	int	Higher priority rules run first

Condition fields: merchant_normalized, description, category, bank_name, amount, transaction_type Condition operators: contains, equals, starts_with, gt, lt, not_equals Actions: set_category, set_merchant, add_tag_ids, apply_split

---

split_payments

Records of actual cash settlements between participants.

Column	Type	Description
from_participant_id	int FK → participants	Who paid
to_participant_id	int FK → participants	Who received
amount	numeric	Amount settled
payment_date	date	Date of settlement
notes	text	Optional note (e.g. "bank transfer")
linked_transaction_id	int FK → transactions (nullable)	If the payment was itself a transaction

---

expense_metadata

Enrichment records for non-statement expenses (email receipts, manual entries). Linked to a transaction if one exists; otherwise a standalone record awaiting reconciliation.

Column	Type	Description
transaction_id	int FK → transactions (unique, nullable)	Linked transaction; NULL until reconciled
source	text	Origin: email, manual
paperless_doc_id	int	Paperless-NGX document ID
payment_method	text	credit_card, debit_card, paypal, afterpay, cash, etc.
payment_method_detail	text	Card last-4 or provider detail
order_reference	text	Order/confirmation number
line_items	jsonb	Array of {description, qty, unit_price, total}
merchant_normalized	text	Canonical merchant for matching
amount / transaction_date	numeric / date	Used for reconciliation matching when transaction_id IS NULL
extraction_model	text	AI model used (gemini-2.5-flash)

Partial index on (merchant_normalized, transaction_date) WHERE transaction_id IS NULL powers reconciliation queries.

---

rule_apply_runs

Audit log of bulk rule-apply operations. Each run captures which transactions were affected and a full snapshot for revert support.

Column	Type	Description
owner_id	int FK → participants
applied_at	timestamptz	When the run executed
split_from	date	Optional date filter used for this run
matched	int	Number of rules matched
transactions_affected	int	Number of transactions changed
reverted_at	timestamptz	Set when run was reverted
snapshot	jsonb	Pre-run state of all affected transactions

---

budgets

Monthly spend targets per category. Stored but currently unused in the UI (replaced by the analytics/insights views).

Column	Type	Description
owner_id	int FK → participants
category	text	Category name
month	date	Always first of month (e.g. 2026-03-01)
amount_limit	numeric	Spend target for that category/month

---

Category Taxonomy

Fixed set defined in src/lib/categories.ts. Applied by Gemini at ingestion and overridable by the user or rules engine:

groceries · dining · transport · fuel · shopping · utilities · entertainment · travel · health · insurance · subscriptions · cash_advance · government · education · rent · home_goods · home_maintenance · transfers · income · investment · personal_care · pets · gifts · charity · other

• home_goods — items purchased for the house (appliances, furniture, kitchenware, electronics)
• home_maintenance — services on the property (cleaning, mowing, repairs)

Committed spend (Insights page): rent, utilities, insurance, subscriptions Excluded from spend analytics: transfers, investment

---

API Routes

All routes require authentication via X-Forwarded-User header (set by Traefik). Responses are always scoped to the authenticated user's owner_id.

Method	Route	Description
GET	/api/statements	All statements for current user
GET / PATCH	/api/statements/[id]	Get statement; PATCH to reassign owner (also writes account_owner_mappings)
GET	/api/transactions	Paginated transactions. Filters: from, to, categories, bank_names, tag_ids, transaction_types, search, statement_id, amount_min, amount_max, has_split (yes/no). Sort: sort_by (transaction_date|amount|created_at), sort_dir (asc|desc)
POST	/api/transactions	Create a manual transaction (no statement)
GET / PATCH	/api/transactions/[id]	Get transaction; PATCH to upsert override (category, merchant, notes)
GET / POST	/api/transactions/[id]/splits	List or create splits on a transaction
GET / POST	/api/transactions/[id]/tags	List or apply tags to a transaction
POST	/api/transactions/bulk	Bulk update category/merchant across multiple transactions
POST	/api/transactions/reconcile	Link manual transactions to statement transactions; copies overrides, tags, splits across
GET	/api/analytics/monthly	Split-adjusted monthly spend by category + income + investments. Params: months (1–24, default 6)
GET	/api/analytics/subscriptions	Recurring charge detection — merchants with ≥3 occurrences at consistent intervals
GET	/api/analytics/fees	Fees and interest from statement summaries + individual fee/interest transactions
GET	/api/shared-transactions	Transactions with active splits; sorted client-side by date/imported/amount in the UI
POST	/api/splits/settle	Mark a split as settled
GET / POST	/api/split-payments	List or record cash settlements between participants
GET / POST	/api/participants	List participants; POST to create (with optional email)
GET	/api/participants/[id]/balance	Net balance owed by/to a specific participant
GET	/api/participants/balances	All participant balances
GET / POST	/api/rules	List or create rules
PATCH / DELETE	/api/rules/[id]	Update or delete a rule
POST	/api/rules/apply	Run all enabled rules against all transactions; returns {matched, transactions_affected}
GET / POST	/api/budgets	List budgets for a month (?month=YYYY-MM); upsert budget
DELETE	/api/budgets/[id]	Delete a budget
GET	/api/merchants	Merchant name autocomplete suggestions
GET	/api/me	Current user info derived from X-Forwarded-User header
GET / POST	/api/tags	List or create tags
PATCH / DELETE	/api/tags/[id]	Update or delete a tag

Machine routes — shared-secret x-ingest-token header instead of X-Forwarded-User; each fails closed when its token is unset:

Method	Route	Token	Description
POST / PATCH	/api/orders/ingest	ORDER_INGEST_TOKEN	n8n order receipts → transactions (see smarthome CLAUDE.md "Order Processor")
POST	/api/receipts/ingest	RECEIPT_INGEST_TOKEN	pantry-app confirmed grocery receipts → one transaction per tender leg (ING-10)
POST	/api/intake	INTAKE_TOKEN	Intake router (2026-08-21): multipart source (uber_eats|doordash|coles|woolworths|other), shared_by, batch_id (uuid), 1–6 files (image / PDF / text). Persists intake_batches, extracts once via LiteLLM cloud-budget (PDF → text via unpdf), matches (platform, merchant, date, total) exactly, then enriches the matched order's line_items (food), forwards the original files to pantry POST /api/capture (grocery), or replies with candidates / "Groceries or a meal?" / needs-header / held / duplicate. Returns {kind, status, transaction_id?, capture_id?, candidates?, slack_blocks, text}
PUT	/api/transactions/[id]/line-items	INTAKE_TOKEN or RECEIPT_INGEST_TOKEN	Idempotent replace of expense_metadata.line_items for transaction_id = :id OR matched_transaction_id = :id; body {line_items, source: 'intake'|'pantry', batch_id?, platform?, replace: true}; creates the metadata row if absent, never a transaction (404). Returns {previous_line_items, line_items} — pantry calls it on every confirm and with previous_line_items on undo
POST	/api/slack/interactive	ORDER_INGEST_TOKEN + Slack signature	Order-nudge buttons and the intake taps (intake_use_candidate, intake_create, intake_grocery, intake_meal)

Env for the intake router: INTAKE_TOKEN, LITELLM_BASE_URL + LITELLM_KEY (OpenAI-compatible, alias cloud-budget), PANTRY_BASE_URL (+ the existing RECEIPT_INGEST_TOKEN as pantry's machine auth), optional INTAKE_FILE_DIR (original files by sha256, default /app/data/intake — a later "Groceries" tap re-sends them) and FINANCE_PUBLIC_URL (the card's View link, default https://finance.bosecamp.com). Design: smarthome plans/intake-router-slice-plan.md §11.

---

Ingestion Pipeline

Paperless-NGX
  └─ documents tagged "Bank Statement" + "Credit Card" (without "cc-processor")
       │
       ▼
  N8N workflow — polls every 5 minutes (workflow ID: FysADdFwEtwONQl4)
       │
       ├─ Duplicate check: SELECT WHERE paperless_doc_id = <id>
       │    └─ Already processed → skip, mark in Paperless
       │
       ├─ Download PDF binary from Paperless API
       │
       ├─ Gemini 2.5 Flash — PDF → structured JSON
       │    responseSchema: { summary: {...}, transactions: [...] }
       │    timeout: 180s, retryOnFail: 3×, delay: 30s
       │
       ├─ Parse & normalise
       │    account_number: strip spaces
       │    bank_name: title-case
       │    FX rate: fetch live from open.er-api.com if non-AUD
       │
       ├─ Statement exists? (bank + account + billing_end_date)
       │    └─ Duplicate → skip, mark in Paperless
       │
       ├─ New bank? → Slack approval gate (human confirms before insert)
       │
       ├─ Lookup account_owner_mappings → resolve owner_id (default: 1 = "Me")
       │
       ├─ INSERT statements + transactions
       │
       ├─ Google Calendar reminder for payment_due_date (credit cards)
       │
       └─ Paperless: PATCH document to add "cc-processor" tag

N8N workflow JSON: docker/automation/workflows/cc-statement-processor-paperless.json in the smarthome repo.

---

Schema Migrations

Located in prisma/migrations/. Applied manually against the running container:

docker exec postgres-personal psql -U personal -d personal \
  < prisma/migrations/<migration>/migration.sql

Migration	What it adds
0001_init	statements, transactions, participants
0002_splits	transaction_splits
0003_owner_segregation	owner_id on statements, account_owner_mappings, email on participants
0004_tags	tags, transaction_tags
0005_rules	rules
0006_budgets	budgets
0007_cashflow	amount_aud, exchange_rate_to_aud on transactions; exchange_rate_to_aud on statements
0008_my_share_percent	my_share_percent on transaction_overrides
0009_split_payments	split_payments
0010_csv_import_reconcile	reconciled_with_id, CSV import support
0011_trips	trips, trip_id on transaction_overrides (catch-up — was applied directly)
…
0034_intake_batches	intake_batches (one row per /api/intake upload, written before extraction) and intake_merchant_kinds (the remembered "Groceries or a meal?" answer per merchant)

paperless_doc_id on statements and the uq_statements_paperless_doc_id index were added directly (not tracked in a migration file). owner_id on transactions and statement_id made nullable were applied directly (March 2026) to support manual transaction entry without a fake statement. reconciled_with_id on transactions, expense_metadata, rule_apply_runs, split_payments were added directly and are covered by the Prisma schema but lack individual migration files.

---

Known Gaps / TODOs

Payment Provider tracking

Currently merchant_normalized conflates the payment provider with the merchant. Transactions processed through PayPal, Afterpay, Zip, Alipay, etc. end up with the provider as the merchant when the real merchant can't be recovered.

What's been done so far:

• PayPal entries that embed the merchant name (e.g. PAYPAL *BUNNINGSGRO) were cleaned up — the real merchant was extracted during the March 2026 consolidation pass.
• Pure PayPal/Afterpay/Zip entries where the merchant is unrecoverable were left as-is.
• A one-time SQL consolidation pass normalised ~50 merchant name variant groups (March 2026).

Remaining work:

1. DB migration: ALTER TABLE transactions ADD COLUMN payment_provider text and same on transaction_overrides.
2. Gemini prompt: add payment_provider to the responseSchema so the AI extracts it separately ("PayPal", "Afterpay", "Zip", null, etc.) — the raw bank description usually contains enough signal.
3. Backfill: for existing transactions, derive payment_provider from merchant_name patterns (PAYPAL *, AFTERPAY, ZIP/ZIPPAY, BPAY).
4. App: surface payment_provider as a filter/column in the transactions view; exclude payment providers from merchant analytics so they don't inflate the merchant list.

---

Agent Access (MCP)

LLM agents read this data through the read-only postgres-personal MCP server (7 typed finance tools + guarded adhoc_query, agent_ro role, 100-row cap, mcp_query_log audit). It's a stdio server, so any MCP client can use it — Hermes, Claude Code, Codex, Claude Desktop — locally via docker exec or remotely over SSH.

→ docs/agent-access.md — architecture, tool + view reference, per-client setup, and what changing the schema means for the agent views.

---

Deployment

Runs as a Docker container alongside the rest of the home lab stack.

Push-to-deploy (default, since 2026-07-19): pushing to main on Gitea fires a webhook to Komodo's deploy-finance Procedure, which redeploys the finance stack with --build. CI (Gitea Actions) runs lint + unit tests on the same push.

Manual fallback if Komodo is unavailable:

# From smarthome repo root
docker compose --env-file docker/common.env --env-file docker/finance/.env \
  -f docker/finance/docker-compose.yml up -d --build

The container is only reachable via Traefik (https://finance.bosecamp.com, forward-auth sets X-Forwarded-User) plus a 127.0.0.1-bound host port for on-box debugging — the app trusts the X-Forwarded-User header, so it must never be directly reachable from the LAN.

The container uses Next.js standalone output. @prisma/adapter-pg and pg are listed in serverExternalPackages in next.config.ts to ensure they are included in the standalone bundle.