${title} — Spend Agent

View production setup (real banks)

Sign up at teller.io and create an application — copy the Application ID.
Apply for production access in the Teller dashboard (free up to 100 accounts, requires ID verification).
Download certificate.pem + private_key.pem — these mTLS certs authenticate every API call.
In Val Town env vars: set TELLER_APP_ID, TELLER_ENV=production, TELLER_CERT, TELLER_KEY (paste full PEMs).
Reload and click Connect Bank.

PEMs stay server-side as env vars. Teller production docs →

`\n\n// ── PEM Reconstruction ──────────────────────────────────────────\n// Val Town env vars strip newlines. This rebuilds valid PEM from flattened input.\nexport function fixPem(raw: string): string {\n if (raw.includes('\\n')) return raw\n const beginMatch = raw.match(/-----BEGIN ([A-Z ]+)-----/)\n const endMatch = raw.match(/-----END ([A-Z ]+)-----/)\n if (!beginMatch || !endMatch) return raw\n const type = beginMatch[1]\n let b64 = raw\n .replace(`-----BEGIN ${type}-----`, '')\n .replace(`-----END ${type}-----`, '')\n .replace(/\\s+/g, '')\n const lines: string[] = []\n for (let i = 0; i < b64.length; i += 64) {\n lines.push(b64.slice(i, i + 64))\n }\n return `-----BEGIN ${type}-----\\n${lines.join('\\n')}\\n-----END ${type}-----\\n`\n}\n\n// ── Category Colors ─────────────────────────────────────────────\nexport const CAT_COLORS: Record = {\n Housing: '#8b5cf6',\n Food: '#f97316',\n Transportation: '#3b82f6',\n Shopping: '#a855f7',\n Entertainment: '#ec4899',\n Health: '#ef4444',\n Subscriptions: '#6366f1',\n Bills: '#64748b',\n Travel: '#14b8a6',\n Investments: '#22d3ee',\n Transfers: '#475569',\n Salary: '#059669',\n Income: '#10b981',\n Gifts: '#f59e0b',\n Other: '#6b7280',\n}\n\n// ── HTML Shell ──────────────────────────────────────────────────\nexport function pageShell(title: string, activePath: string, body: string, scripts = ''): string {\n const nav = [\n { path: '/', label: 'Overview' },\n { path: '/ask', label: 'Ask' },\n { path: '/settings', label: 'Settings' },\n ]\n .map((n) => {\n const active = activePath === n.path\n return `${n.label}`\n })\n .join('')\n\n return `\n\n\n \n ${title} — Spend Agent\n \n \n\n\n

Spend Agent

${nav}

\n ${body}\n

\n ${scripts}\n\n`\n}\n",[146],"2abace02-5acf-11f1-ba12-ee650bb23af1","db.ts","89fb741c-5aab-11f1-8905-ee650bb23af1",200,["D",1779998124644],6,"// Spend Agent — Database Schema & Helpers\n//\n// Uses the val-scoped SQLite database (`std/sqlite/main.ts`) — each val gets its\n// own isolated database, so plain unprefixed table names are safe (no collision\n// with other vals). The legacy account-global `std/sqlite` is shared across all\n// vals and is intentionally NOT used here.\nimport sqlite from 'https://esm.town/v/std/sqlite/main.ts'\nimport { CAT_COLORS } from './lib.ts'\n\nexport { sqlite }\n\n// Canonical table names — referenced by name from other modules so a future\n// rename only touches this file.\nexport const T = {\n tx: 'transactions',\n summaries: 'monthly_summaries',\n config: 'config',\n tellerTokens: 'teller_tokens',\n} as const\n\n// ── Schema ──────────────────────────────────────────────────────\nexport async function ensureSchema() {\n await sqlite.batch([\n `CREATE TABLE IF NOT EXISTS transactions (\n id TEXT PRIMARY KEY,\n source TEXT NOT NULL,\n source_id TEXT,\n date TEXT NOT NULL,\n description TEXT NOT NULL,\n amount REAL NOT NULL,\n abs_amount REAL NOT NULL,\n account_name TEXT,\n institution TEXT,\n category TEXT,\n category_source TEXT,\n llm_confidence REAL,\n llm_reasoning TEXT,\n is_transfer INTEGER DEFAULT 0,\n is_income INTEGER DEFAULT 0,\n is_duplicate INTEGER DEFAULT 0,\n duplicate_of TEXT,\n status TEXT DEFAULT 'posted',\n created_at TEXT DEFAULT (datetime('now'))\n )`,\n `CREATE TABLE IF NOT EXISTS monthly_summaries (\n month TEXT PRIMARY KEY,\n total_income REAL,\n total_spend REAL,\n total_transfers REAL,\n net_cash_flow REAL,\n category_breakdown TEXT,\n tx_count INTEGER\n )`,\n `CREATE TABLE IF NOT EXISTS config (\n key TEXT PRIMARY KEY,\n value TEXT\n )`,\n `CREATE TABLE IF NOT EXISTS teller_tokens (\n id TEXT PRIMARY KEY,\n access_token TEXT NOT NULL,\n enrollment_id TEXT,\n institution TEXT NOT NULL,\n linked_at TEXT DEFAULT (datetime('now'))\n )`,\n ])\n}\n\n// ── Config helpers ──────────────────────────────────────────────\nexport async function getConfig(key: string, fallback = ''): Promise {\n const res = await sqlite.execute({\n sql: 'SELECT value FROM config WHERE key = ?',\n args: [key],\n })\n return res.rows.length > 0 ? String(res.rows[0][0] ?? fallback) : fallback\n}\n\nexport async function setConfig(key: string, value: string) {\n await sqlite.execute({\n sql: 'INSERT OR REPLACE INTO config (key, value) VALUES (?, ?)',\n args: [key, value],\n })\n}\n\n// ── Monthly summary recompute ───────────────────────────────────\nexport async function recomputeMonth(month: string) {\n const stats = await sqlite.execute({\n sql: `SELECT\n SUM(CASE WHEN is_income = 1 AND is_transfer = 0 THEN abs_amount ELSE 0 END),\n SUM(CASE WHEN is_income = 0 AND is_transfer = 0 AND is_duplicate = 0 THEN abs_amount ELSE 0 END),\n SUM(CASE WHEN is_transfer = 1 THEN abs_amount ELSE 0 END),\n COUNT(*)\n FROM transactions WHERE substr(date, 1, 7) = ?`,\n args: [month],\n })\n const r = stats.rows[0] as any[]\n const income = Number(r[0]) || 0\n const spend = Number(r[1]) || 0\n const transfers = Number(r[2]) || 0\n const txCount = Number(r[3]) || 0\n\n const catRes = await sqlite.execute({\n sql: `SELECT category, SUM(abs_amount) FROM transactions\n WHERE substr(date, 1, 7) = ? AND is_income = 0 AND is_transfer = 0 AND is_duplicate = 0\n GROUP BY category ORDER BY SUM(abs_amount) DESC`,\n args: [month],\n })\n const cats: Record = {}\n for (const cr of catRes.rows)\n cats[String(cr[0] || 'Other')] = Math.round(Number(cr[1]) * 100) / 100\n\n await sqlite.execute({\n sql: `INSERT OR REPLACE INTO monthly_summaries (month, total_income, total_spend, total_transfers, net_cash_flow, category_breakdown, tx_count)\n VALUES (?, ?, ?, ?, ?, ?, ?)`,\n args: [\n month,\n Math.round(income * 100) / 100,\n Math.round(spend * 100) / 100,\n Math.round(transfers * 100) / 100,\n Math.round((income - spend) * 100) / 100,\n JSON.stringify(cats),\n txCount,\n ],\n })\n}\n\n// ── Default system prompt ───────────────────────────────────────\nexport const DEFAULT_SYSTEM_PROMPT = `You are a personal finance categorizer. Classify each transaction into exactly one category.\n\nCategories are enforced by schema — focus on rules for ambiguous cases:\n\nRules:\n- Venmo/Zelle payments TO someone are the actual spend category, not \"Transfers\"\n- Venmo/Zelle payments FROM someone are income or reimbursements\n- Credit card payments and bank transfers between own accounts are \"Transfers\"\n- Recurring charges (Netflix, Spotify, gym) are \"Subscriptions\"\n- Restaurant and food delivery (DoorDash, UberEats, Grubhub) are \"Food\"\n- Uber/Lyft rides are \"Transportation\"\n- Amazon purchases are \"Shopping\" unless clearly groceries\n\nIncome/credit classification:\n- PAYROLL, direct deposits from employers → \"Salary\"\n- Venmo/Zelle payments FROM someone for rent → \"Housing\"\n- Reimbursements: use the category of what was reimbursed (e.g., trip reimbursement → \"Travel\", utility reimbursement → \"Bills\")\n- Tax refunds (IRS, state) → \"Income\"\n- Credit card statement credits, hotel credits, rewards → use underlying category (travel credit → \"Travel\", entertainment credit → \"Entertainment\")\n- PayPal/Cash App transfers to yourself → \"Transfers\"\n- Family gifts via Zelle/Venmo → \"Gifts\"\n- Other misc income → \"Income\"`\n\n// ── Categories ──────────────────────────────────────────────────\nexport const DEFAULT_CATEGORIES = [\n 'Housing',\n 'Food',\n 'Transportation',\n 'Shopping',\n 'Entertainment',\n 'Health',\n 'Subscriptions',\n 'Bills',\n 'Travel',\n 'Investments',\n 'Transfers',\n 'Salary',\n 'Income',\n 'Gifts',\n 'Other',\n] as const\n\n// Keep CATEGORIES as alias for backward compat\nexport const CATEGORIES = DEFAULT_CATEGORIES\n\nexport type Category = (typeof DEFAULT_CATEGORIES)[number]\n\n// Get user-customized categories (falls back to defaults)\nexport async function getCategories(): Promise {\n const stored = await getConfig('categories', '')\n if (stored) {\n try {\n return JSON.parse(stored) as string[]\n } catch {}\n }\n return [...DEFAULT_CATEGORIES]\n}\n\nexport async function setCategories(cats: string[]) {\n // Always ensure \"Other\" exists as fallback\n if (!cats.includes('Other')) cats.push('Other')\n await setConfig('categories', JSON.stringify(cats))\n}\n\n// Read user-set category colors (overlays CAT_COLORS defaults).\nexport async function getCategoryColors(): Promise> {\n const raw = await getConfig('category_colors', '')\n if (!raw) return {}\n try {\n return JSON.parse(raw) as Record\n } catch {\n return {}\n }\n}\n\nexport async function setCategoryColors(colors: Record) {\n await setConfig('category_colors', JSON.stringify(colors))\n}\n\n// Merge user-saved colors into the in-process CAT_COLORS map. Called once per\n// request so the dashboard and Settings UI both see the current palette.\nexport async function loadCategoryColors(): Promise {\n const userColors = await getCategoryColors()\n for (const [name, hex] of Object.entries(userColors)) {\n if (hex) CAT_COLORS[name] = hex\n }\n}\n\nexport type CategoryRow = { original: string; current: string; color?: string }\n\nexport type SaveCategoriesResult = {\n added: string[]\n renamed: { from: string; to: string; count: number }[]\n removed: string[]\n unclassifiedCount: number\n saved: string[]\n colors: Record\n}\n\nexport async function saveCategoriesAndCleanup(rows: CategoryRow[]): Promise {\n // Normalize each row, dedupe by current name (case-sensitive), drop empty\n const seen = new Set()\n const clean: CategoryRow[] = []\n for (const r of rows || []) {\n const current = (r?.current || '').trim()\n if (!current || seen.has(current)) continue\n seen.add(current)\n clean.push({\n original: (r?.original || '').trim(),\n current,\n color: r?.color,\n })\n }\n\n const oldCats = await getCategories()\n const oldColors = await getCategoryColors()\n const newNames = new Set(clean.map((r) => r.current))\n\n // Anything in old that doesn't appear as anyone's \"original\" is REMOVED.\n // (Rename rows have original=oldName so the oldName isn't considered removed.)\n const claimedOriginals = new Set(clean.map((r) => r.original).filter(Boolean))\n const removed = oldCats.filter(\n (c) => c !== 'Other' && !newNames.has(c) && !claimedOriginals.has(c)\n )\n\n const added: string[] = []\n const renamed: { from: string; to: string; count: number }[] = []\n const newColors: Record = {}\n\n for (const row of clean) {\n if (row.color) newColors[row.current] = row.color\n if (!row.original) {\n // Brand-new category\n if (!oldCats.includes(row.current)) added.push(row.current)\n continue\n }\n if (row.original === row.current) continue // kept as-is\n // Rename: update transactions to the new name\n const countRes = await sqlite.execute({\n sql: 'SELECT COUNT(*) FROM transactions WHERE category = ?',\n args: [row.original],\n })\n const count = Number((countRes.rows[0] as any[])[0]) || 0\n if (count > 0) {\n await sqlite.execute({\n sql: 'UPDATE transactions SET category = ? WHERE category = ?',\n args: [row.current, row.original],\n })\n }\n renamed.push({ from: row.original, to: row.current, count })\n }\n\n // Unclassify transactions tied to removed categories\n let unclassifiedCount = 0\n for (const cat of removed) {\n const countRes = await sqlite.execute({\n sql: 'SELECT COUNT(*) FROM transactions WHERE category = ?',\n args: [cat],\n })\n const n = Number((countRes.rows[0] as any[])[0]) || 0\n if (n === 0) continue\n await sqlite.execute({\n sql: `UPDATE transactions\n SET category = NULL, category_source = NULL,\n llm_confidence = NULL, llm_reasoning = NULL\n WHERE category = ?`,\n args: [cat],\n })\n unclassifiedCount += n\n }\n\n // Persist new categories list (current names)\n await setCategories(clean.map((r) => r.current))\n\n // Persist new colors (drop entries for removed categories, propagate renames)\n const finalColors: Record = {}\n for (const r of clean) {\n const c = newColors[r.current] || oldColors[r.original] || oldColors[r.current]\n if (c) finalColors[r.current] = c\n }\n await setCategoryColors(finalColors)\n // Reflect into the in-process map so subsequent renders see it\n for (const [name, hex] of Object.entries(finalColors)) CAT_COLORS[name] = hex\n for (const cat of removed) delete CAT_COLORS[cat]\n\n const saved = await getCategories()\n return {\n added,\n renamed,\n removed,\n unclassifiedCount,\n saved,\n colors: finalColors,\n }\n}\n\n// ── Demo seed data (template only) ──────────────────────────────\n//\n// Inserts ~36 realistic transactions across the last 3 calendar months so a\n// remixer sees a populated dashboard on first load. No-op if the table already\n// has rows. Dates are computed from `now` so the demo always looks fresh.\n//\n// Most rows are pre-classified (`category_source = 'seed'`); a few are left\n// unclassified so the \"Review & Classify\" banner appears and the remixer can\n// experience the LLM classification flow on first visit.\n\ntype SeedRow = {\n monthOffset: number // 0 = current month, 1 = last, 2 = two ago\n day: number\n description: string\n amount: number // negative = spend, positive = credit\n account: string\n institution: string\n category: string | null // null leaves it for the classifier\n isIncome?: 0 | 1\n}\n\nconst SEED_ROWS: SeedRow[] = [\n // Current month\n {\n monthOffset: 0,\n day: 1,\n description: 'PAYROLL DIRECT DEPOSIT — ACME CORP',\n amount: 5800.0,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Salary',\n isIncome: 1,\n },\n {\n monthOffset: 0,\n day: 1,\n description: 'RENT — 123 MAIN ST APARTMENTS',\n amount: -2400.0,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Housing',\n },\n {\n monthOffset: 0,\n day: 2,\n description: 'NETFLIX.COM',\n amount: -15.49,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Subscriptions',\n },\n {\n monthOffset: 0,\n day: 3,\n description: 'WHOLE FOODS MARKET',\n amount: -87.43,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n {\n monthOffset: 0,\n day: 4,\n description: 'UBER TRIP',\n amount: -18.2,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Transportation',\n },\n {\n monthOffset: 0,\n day: 5,\n description: 'DOORDASH',\n amount: -34.5,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n {\n monthOffset: 0,\n day: 6,\n description: 'SPOTIFY USA',\n amount: -10.99,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Subscriptions',\n },\n {\n monthOffset: 0,\n day: 7,\n description: 'AMAZON.COM',\n amount: -42.18,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Shopping',\n },\n {\n monthOffset: 0,\n day: 8,\n description: 'SHELL GAS STATION',\n amount: -52.3,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Transportation',\n },\n {\n monthOffset: 0,\n day: 9,\n description: \"TRADER JOE'S\",\n amount: -64.12,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n {\n monthOffset: 0,\n day: 10,\n description: 'AMC THEATER',\n amount: -28.5,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: null,\n },\n {\n monthOffset: 0,\n day: 11,\n description: 'VENMO PAYMENT — REIMBURSEMENT FROM SAM',\n amount: 45.0,\n account: 'Venmo',\n institution: 'Venmo',\n category: null,\n isIncome: 1,\n },\n\n // Last month\n {\n monthOffset: 1,\n day: 1,\n description: 'PAYROLL DIRECT DEPOSIT — ACME CORP',\n amount: 5800.0,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Salary',\n isIncome: 1,\n },\n {\n monthOffset: 1,\n day: 1,\n description: 'RENT — 123 MAIN ST APARTMENTS',\n amount: -2400.0,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Housing',\n },\n {\n monthOffset: 1,\n day: 2,\n description: 'NETFLIX.COM',\n amount: -15.49,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Subscriptions',\n },\n {\n monthOffset: 1,\n day: 3,\n description: 'WHOLE FOODS MARKET',\n amount: -112.88,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n {\n monthOffset: 1,\n day: 5,\n description: 'UBER EATS',\n amount: -28.45,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n {\n monthOffset: 1,\n day: 7,\n description: 'AMAZON.COM',\n amount: -156.22,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Shopping',\n },\n {\n monthOffset: 1,\n day: 8,\n description: 'SPOTIFY USA',\n amount: -10.99,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Subscriptions',\n },\n {\n monthOffset: 1,\n day: 10,\n description: 'CVS PHARMACY',\n amount: -18.34,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Health',\n },\n {\n monthOffset: 1,\n day: 12,\n description: 'DELTA AIRLINES',\n amount: -342.5,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Travel',\n },\n {\n monthOffset: 1,\n day: 14,\n description: 'HILTON HOTEL',\n amount: -218.0,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Travel',\n },\n {\n monthOffset: 1,\n day: 16,\n description: \"TRADER JOE'S\",\n amount: -71.9,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n {\n monthOffset: 1,\n day: 18,\n description: 'VENMO TO MOM — GIFT',\n amount: -100.0,\n account: 'Venmo',\n institution: 'Venmo',\n category: 'Gifts',\n },\n {\n monthOffset: 1,\n day: 20,\n description: 'SHELL GAS STATION',\n amount: -48.75,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Transportation',\n },\n {\n monthOffset: 1,\n day: 22,\n description: 'COMCAST INTERNET',\n amount: -89.99,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Bills',\n },\n {\n monthOffset: 1,\n day: 25,\n description: 'VANGUARD INVEST',\n amount: -500.0,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Investments',\n },\n {\n monthOffset: 1,\n day: 28,\n description: 'CHIPOTLE',\n amount: -14.2,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n\n // Two months ago\n {\n monthOffset: 2,\n day: 1,\n description: 'PAYROLL DIRECT DEPOSIT — ACME CORP',\n amount: 5800.0,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Salary',\n isIncome: 1,\n },\n {\n monthOffset: 2,\n day: 1,\n description: 'RENT — 123 MAIN ST APARTMENTS',\n amount: -2400.0,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Housing',\n },\n {\n monthOffset: 2,\n day: 2,\n description: 'NETFLIX.COM',\n amount: -15.49,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Subscriptions',\n },\n {\n monthOffset: 2,\n day: 4,\n description: 'WHOLE FOODS MARKET',\n amount: -98.22,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n {\n monthOffset: 2,\n day: 6,\n description: 'LYFT',\n amount: -22.4,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Transportation',\n },\n {\n monthOffset: 2,\n day: 8,\n description: 'SPOTIFY USA',\n amount: -10.99,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Subscriptions',\n },\n {\n monthOffset: 2,\n day: 10,\n description: 'AMAZON.COM',\n amount: -78.15,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Shopping',\n },\n {\n monthOffset: 2,\n day: 12,\n description: \"TRADER JOE'S\",\n amount: -55.3,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n {\n monthOffset: 2,\n day: 15,\n description: 'VENMO FROM ALEX — DINNER REIMBURSEMENT',\n amount: 32.5,\n account: 'Venmo',\n institution: 'Venmo',\n category: 'Food',\n isIncome: 1,\n },\n {\n monthOffset: 2,\n day: 18,\n description: 'EQUINOX MEMBERSHIP',\n amount: -185.0,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: null,\n },\n {\n monthOffset: 2,\n day: 20,\n description: 'PG&E ELECTRIC',\n amount: -94.2,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Bills',\n },\n {\n monthOffset: 2,\n day: 22,\n description: 'STARBUCKS',\n amount: -6.75,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: 'Food',\n },\n {\n monthOffset: 2,\n day: 25,\n description: 'VANGUARD INVEST',\n amount: -500.0,\n account: 'Chase Checking',\n institution: 'Chase',\n category: 'Investments',\n },\n {\n monthOffset: 2,\n day: 27,\n description: 'TARGET',\n amount: -67.88,\n account: 'Chase Sapphire',\n institution: 'Chase',\n category: null,\n },\n]\n\nfunction monthString(offset: number): string {\n const now = new Date()\n const d = new Date(now.getFullYear(), now.getMonth() - offset, 1)\n return `${d.getFullYear()}-${String(d.getMonth() + 1).padStart(2, '0')}`\n}\n\n// Realistic merchant pool for the \"Simulate 10 transactions\" demo button.\n// Category is left NULL so simulated rows flow through the normal AI\n// classification loop (the button triggers /classify right after).\nconst SIM_MERCHANTS: { desc: string; min: number; max: number; income?: boolean }[] = [\n { desc: 'WHOLEFDS MKT #10234', min: 28, max: 142 },\n { desc: 'TRADER JOES #455', min: 22, max: 88 },\n { desc: 'STARBUCKS STORE 07731', min: 4, max: 17 },\n { desc: 'CHIPOTLE 1845', min: 11, max: 29 },\n { desc: 'UBER *TRIP', min: 8, max: 41 },\n { desc: 'LYFT *RIDE WED 8PM', min: 7, max: 38 },\n { desc: 'SHELL OIL 57442109', min: 32, max: 74 },\n { desc: 'AMAZON.COM*RT4U91PB3', min: 14, max: 156 },\n { desc: 'NETFLIX.COM', min: 15, max: 23 },\n { desc: 'SPOTIFY USA', min: 11, max: 11 },\n { desc: 'DELTA AIR 0062371882', min: 184, max: 612 },\n { desc: 'BLUE BOTTLE COFFEE', min: 5, max: 19 },\n { desc: 'CVS/PHARMACY #8821', min: 9, max: 67 },\n { desc: 'EQUINOX FITNESS', min: 215, max: 215 },\n { desc: 'PG&E WEB ONLINE', min: 48, max: 132 },\n { desc: 'DOORDASH*SWEETGREEN', min: 18, max: 44 },\n { desc: 'APPLE.COM/BILL', min: 3, max: 29 },\n { desc: 'PAYROLL DIRECT DEP', min: 3200, max: 3200, income: true },\n]\n\n// Insert 10 realistic, unclassified transactions spread over the last 28 days.\n// Reuses an existing linked account name if present. Returns count inserted.\nexport async function insertSimulatedTransactions(): Promise {\n await ensureSchema()\n const acctRes = await sqlite.execute(\n `SELECT account_name, institution FROM ${T.tx} WHERE account_name IS NOT NULL AND account_name != '' ORDER BY date DESC LIMIT 1`\n )\n const acctName = String((acctRes.rows[0]?.[0] as any) || 'Everyday Checking')\n const institution = String((acctRes.rows[0]?.[1] as any) || 'Simulated Bank')\n\n const pool = [...SIM_MERCHANTS].sort(() => Math.random() - 0.5).slice(0, 10)\n const affected = new Set()\n let inserted = 0\n for (let i = 0; i < pool.length; i++) {\n const m = pool[i]\n const amt = Math.round((m.min + Math.random() * (m.max - m.min)) * 100) / 100\n const d = new Date()\n d.setDate(d.getDate() - Math.floor(Math.random() * 28))\n const date = d.toISOString().slice(0, 10)\n const isIncome = m.income ? 1 : 0\n const amount = isIncome ? amt : -amt\n await sqlite.execute({\n sql: `INSERT INTO ${T.tx} (id, source, source_id, date, description, amount, abs_amount, account_name, institution, is_income, status)\n VALUES (?, 'simulated', ?, ?, ?, ?, ?, ?, ?, ?, 'posted')`,\n args: [\n crypto.randomUUID(),\n `sim-${Date.now()}-${i}`,\n date,\n m.desc,\n amount,\n amt,\n acctName,\n institution,\n isIncome,\n ],\n })\n inserted++\n affected.add(date.slice(0, 7))\n }\n for (const month of affected) await recomputeMonth(month)\n return inserted\n}\n\nexport async function seedDemoData() {\n // One-time cleanup: drop legacy demo teller tokens that earlier template\n // versions auto-seeded. Real Teller-Connect tokens are untouched.\n await sqlite.execute({\n sql: `DELETE FROM ${T.tellerTokens} WHERE id LIKE 'demo_seed_%'`,\n args: [],\n })\n\n const existing = await sqlite.execute({\n sql: `SELECT COUNT(*) FROM ${T.tx}`,\n args: [],\n })\n const count = Number((existing.rows[0] as any[])[0]) || 0\n if (count > 0) return\n\n let i = 0\n for (const row of SEED_ROWS) {\n const month = monthString(row.monthOffset)\n const date = `${month}-${String(row.day).padStart(2, '0')}`\n const id = `seed_${++i}_${date}`\n await sqlite.execute({\n sql: `INSERT INTO ${T.tx} (id, source, date, description, amount, abs_amount, account_name, institution, category, category_source, is_income, is_transfer, status) VALUES (?, 'seed', ?, ?, ?, ?, ?, ?, ?, ?, ?, 0, 'posted')`,\n args: [\n id,\n date,\n row.description,\n row.amount,\n Math.abs(row.amount),\n row.account,\n row.institution,\n row.category,\n row.category ? 'seed' : null,\n row.isIncome ?? 0,\n ],\n })\n }\n\n for (let m = 0; m < 3; m++) await recomputeMonth(monthString(m))\n}\n",[153],"25b57a7a-59f2-11f1-9279-ee650bb23af1","classify.ts","62089d3a-5877-11f1-ab13-ee650bb23af1",86,["D",1779903197447],"// Spend Agent — Daily Classification Cron\n// Thin wrapper around classify-engine.ts for Val Town's interval runner.\nimport { runClassification } from './classify-engine.ts'\n\nexport default async function () {\n const result = await runClassification()\n if (result.errors.length > 0) console.log('Classification errors:', result.errors.join('; '))\n}\n",[161],"eb0117e8-59ee-11f1-bfca-ee650bb23af1","classify-ui.ts","d1610e4a-59eb-11f1-a639-ee650bb23af1",83,["D",1779901810486],3,"// Spend Agent — Classification Drawer UI\n// Extracted to stay under Val Town's 80K char file limit.\nimport { fmt, escHtml, CAT_COLORS } from './lib.ts'\n\ntype UncTx = {\n id: string\n date: string\n description: string\n abs_amount: number\n account_name: string\n institution: string\n}\n\nexport function classifyDrawerHtml(\n uncCount: number,\n uncTxs: UncTx[],\n currentPrompt: string\n): { banner: string; drawer: string; scripts: string } {\n const banner =\n uncCount > 0\n ? `

\n ${uncCount} unclassified transaction${uncCount > 1 ? 's' : ''}\n Review & Classify ▶\n

`\n : ''\n\n const drawer = `\n \n `\n\n const escapedTxs = JSON.stringify(uncTxs)\n const escapedPrompt = JSON.stringify(currentPrompt)\n const escapedColors = JSON.stringify(CAT_COLORS)\n\n const scripts = `\n var uncTxData = ${escapedTxs};\n var previewResults = null;\n var drawerPrompt = ${escapedPrompt};\n var originalPrompt = ${escapedPrompt};\n var catColors = ${escapedColors};\n\n function getCatColor(cat) {\n return catColors[cat] || '#6b7280';\n }\n\n function openClassifyDrawer() {\n document.getElementById('classify-overlay').style.display = 'block';\n document.getElementById('classify-drawer').style.display = 'block';\n document.getElementById('drawer-step-list').style.display = 'block';\n document.getElementById('drawer-step-preview').style.display = 'none';\n document.getElementById('drawer-prompt').value = drawerPrompt;\n renderTxList();\n }\n function closeClassifyDrawer() {\n document.getElementById('classify-overlay').style.display = 'none';\n document.getElementById('classify-drawer').style.display = 'none';\n }\n function onPromptEdit() {\n var btn = document.getElementById('drawer-save-prompt-btn');\n var changed = document.getElementById('drawer-prompt').value !== originalPrompt;\n btn.style.display = changed ? 'inline' : 'none';\n }\n function onPreviewPromptEdit() {\n var btn = document.getElementById('preview-save-prompt-btn');\n var changed = document.getElementById('preview-prompt').value !== originalPrompt;\n btn.style.display = changed ? 'inline' : 'none';\n }\n async function savePrompt() {\n var promptEl = document.getElementById('drawer-prompt') || document.getElementById('preview-prompt');\n var val = promptEl.value;\n var btns = [document.getElementById('drawer-save-prompt-btn'), document.getElementById('preview-save-prompt-btn')];\n btns.forEach(function(b) { if (b) b.textContent = 'Saving...'; });\n try {\n await fetch('/save-prompt', { method: 'POST', headers: {'Content-Type':'application/json'}, body: JSON.stringify({prompt: val}) });\n originalPrompt = val;\n btns.forEach(function(b) { if (b) { b.style.display = 'none'; b.textContent = 'Save'; } });\n // Sync both textareas\n var dp = document.getElementById('drawer-prompt');\n var pp = document.getElementById('preview-prompt');\n if (dp) dp.value = val;\n if (pp) pp.value = val;\n } catch(e) { btns.forEach(function(b) { if (b) b.textContent = 'Error'; }); }\n }\n function renderTxList() {\n var el = document.getElementById('drawer-tx-list');\n if (uncTxData.length === 0) { el.innerHTML = '

No unclassified transactions

'; return; }\n el.innerHTML = uncTxData.map(function(tx) {\n var d = new Date(tx.date + 'T12:00:00').toLocaleDateString('en-US', {month:'short',day:'numeric'});\n return '

' +\n '

' + tx.description.slice(0,45) + '

' +\n '

' + d + ' \\\\u00b7 ' + (tx.account_name || '') + '

' +\n '$' + tx.abs_amount.toFixed(2) + '

';\n }).join('');\n }\n async function runPreview() {\n var btn = document.getElementById('drawer-preview-btn');\n btn.textContent = 'Classifying...'; btn.disabled = true;\n try {\n var promptVal = document.getElementById('drawer-prompt').value;\n var body = {};\n if (promptVal !== originalPrompt) {\n body.prompt = promptVal;\n }\n var res = await fetch('/preview-classify', { method: 'POST', headers: {'Content-Type':'application/json'}, body: JSON.stringify(body) });\n var data = await res.json();\n if (data.error) { btn.textContent = 'Error: ' + data.error; btn.disabled = false; return; }\n previewResults = data.results;\n drawerPrompt = promptVal;\n renderPreview(data);\n document.getElementById('drawer-step-list').style.display = 'none';\n document.getElementById('drawer-step-preview').style.display = 'block';\n document.getElementById('preview-prompt').value = promptVal;\n } catch(e) { alert('Error: ' + e.message); }\n btn.textContent = 'Preview Classification'; btn.disabled = false;\n }\n async function retryFromPreview() {\n var btn = document.getElementById('drawer-retry-btn');\n btn.textContent = 'Classifying...'; btn.disabled = true;\n try {\n var promptVal = document.getElementById('preview-prompt').value;\n var body = {};\n if (promptVal !== originalPrompt) {\n body.prompt = promptVal;\n }\n var res = await fetch('/preview-classify', { method: 'POST', headers: {'Content-Type':'application/json'}, body: JSON.stringify(body) });\n var data = await res.json();\n if (data.error) { btn.textContent = 'Error: ' + data.error; btn.disabled = false; return; }\n previewResults = data.results;\n drawerPrompt = promptVal;\n // Sync prompt back to list step\n document.getElementById('drawer-prompt').value = promptVal;\n renderPreview(data);\n } catch(e) { alert('Error: ' + e.message); }\n btn.textContent = '\\\\u21bb Retry'; btn.disabled = false;\n }\n function renderPreview(data) {\n document.getElementById('preview-count').textContent = data.results.length + ' classified' + (data.errors.length ? ', ' + data.errors.length + ' errors' : '');\n var el = document.getElementById('drawer-preview-list');\n el.innerHTML = data.results.map(function(r) {\n var d = new Date(r.date + 'T12:00:00').toLocaleDateString('en-US', {month:'short',day:'numeric'});\n var conf = r.confidence >= 0.7 ? '#22c55e' : '#f59e0b';\n var catColor = getCatColor(r.category);\n return '

' +\n '

' +\n '' + r.description.slice(0,40) + '' +\n '$' + r.abs_amount.toFixed(2) + '

' +\n '

' +\n '' + r.category + '' +\n '' + Math.round(r.confidence * 100) + '%' +\n '' + (r.reasoning || '').slice(0,60) + '

';\n }).join('');\n }\n function backToList() {\n document.getElementById('drawer-step-list').style.display = 'block';\n document.getElementById('drawer-step-preview').style.display = 'none';\n }\n async function approveResults() {\n if (!previewResults || previewResults.length === 0) return;\n var btn = document.getElementById('drawer-approve-btn');\n btn.textContent = 'Saving...'; btn.disabled = true;\n try {\n var promptEl = document.getElementById('preview-prompt');\n var body = { results: previewResults };\n if (promptEl.value !== originalPrompt) body.prompt = promptEl.value;\n var res = await fetch('/approve-classify', { method: 'POST', headers: {'Content-Type':'application/json'}, body: JSON.stringify(body) });\n var data = await res.json();\n btn.textContent = data.applied + ' applied!';\n setTimeout(function() { location.reload(); }, 1000);\n } catch(e) { btn.textContent = 'Error'; btn.disabled = false; }\n }`\n\n return { banner, drawer, scripts }\n}\n",[168],"bce77fbe-5ae8-11f1-8741-ee650bb23af1","classify-engine.ts","28683b24-5879-11f1-ab12-ee650bb23af1",["D",1780009107326],14,"// Spend Agent — Classification Engine (shared logic)\n// Exported functions used by app.ts (on-demand) and classify.ts (daily cron).\nimport {\n sqlite,\n ensureSchema,\n getConfig,\n DEFAULT_SYSTEM_PROMPT,\n CATEGORIES,\n getCategories,\n recomputeMonth,\n T,\n} from './db.ts'\n// No SDK import — we call the Anthropic Messages API via raw fetch (same as the\n// Ask handler) so nothing heavy is pulled into the val's cold-start path.\n\n// ── Shared tool-use classification helper ───────────────────────\ntype RawTx = {\n id: string\n date: string\n description: string\n amount: number\n abs_amount: number\n account_name: string\n institution: string\n is_income: number\n}\n\ntype ClassifyResult = {\n id: string\n category: string\n confidence: number\n reasoning: string\n}\n\nasync function classifyBatch(\n apiKey: string,\n batch: RawTx[],\n cats: string[],\n systemPrompt: string\n): Promise<{ results: ClassifyResult[]; error?: string }> {\n const txLines = batch\n .map(\n (tx, idx) =>\n `${idx + 1}. [${tx.id}] ${tx.date} | ${tx.description} | $${tx.abs_amount} | ${tx.account_name} (${tx.institution}) | ${tx.is_income ? 'CREDIT/INCOME' : 'DEBIT/SPEND'}`\n )\n .join('\\n')\n\n const tool: any = {\n name: 'classify_transactions',\n description: 'Classify a batch of financial transactions into categories',\n input_schema: {\n type: 'object' as const,\n properties: {\n classifications: {\n type: 'array',\n items: {\n type: 'object',\n properties: {\n id: { type: 'string', description: 'Transaction ID' },\n category: { type: 'string', enum: cats, description: 'Exactly one valid category' },\n confidence: { type: 'number', description: 'Confidence 0-1' },\n reasoning: { type: 'string', description: 'Brief reasoning' },\n },\n required: ['id', 'category', 'confidence', 'reasoning'],\n },\n },\n },\n required: ['classifications'],\n },\n }\n\n const res = await fetch('https://api.anthropic.com/v1/messages', {\n method: 'POST',\n headers: {\n 'x-api-key': apiKey,\n 'anthropic-version': '2023-06-01',\n 'content-type': 'application/json',\n },\n body: JSON.stringify({\n model: 'claude-haiku-4-5-20251001',\n max_tokens: 8192,\n system: systemPrompt,\n tools: [tool],\n tool_choice: { type: 'tool', name: 'classify_transactions' },\n messages: [\n {\n role: 'user',\n content: `Classify these ${batch.length} transactions.\n\nIMPORTANT: For CREDIT/INCOME transactions, classify by underlying purpose — do NOT default to \"Income\":\n- Payroll / direct deposits from employers → \"Salary\"\n- Rent payments from roommate (Venmo/Zelle with \"rent\" in description) → \"Housing\"\n- Travel credits, hotel refunds → \"Travel\"\n- Entertainment credits → \"Entertainment\"\n- Uber/Lyft refunds → \"Transportation\"\n- Utility/wifi reimbursements → \"Bills\"\n- Self-transfers (Cash App to yourself, PayPal transfer) → \"Transfers\"\n- Zelle/Venmo from friends or family (not for rent/bills) → \"Gifts\"\n- Only use \"Income\" for truly miscellaneous income (tax refunds, unknown deposits)\n\nTransactions:\n${txLines}`,\n },\n ],\n }),\n })\n\n if (!res.ok) {\n return { results: [], error: `HTTP ${res.status}: ${(await res.text()).slice(0, 200)}` }\n }\n const data = await res.json()\n const toolBlock = (data.content || []).find((b: any) => b.type === 'tool_use')\n if (!toolBlock) {\n return { results: [], error: 'No tool_use block in response' }\n }\n const input = toolBlock.input as { classifications: ClassifyResult[] }\n return { results: input.classifications }\n}\n\nexport async function runClassification(\n reclassifyAll = false\n): Promise<{ classified: number; skipped: number; deduped: number; errors: string[] }> {\n await ensureSchema()\n const apiKey = Deno.env.get('ANTHROPIC_API_KEY')\n if (!apiKey)\n return { classified: 0, skipped: 0, deduped: 0, errors: ['Missing ANTHROPIC_API_KEY'] }\n\n const errors: string[] = []\n const deduped = await runDedup()\n\n const sql = reclassifyAll\n ? `SELECT id, date, description, amount, abs_amount, account_name, institution, is_income FROM ${T.tx} WHERE category_source != 'user' AND is_duplicate = 0 ORDER BY date DESC`\n : `SELECT id, date, description, amount, abs_amount, account_name, institution, is_income FROM ${T.tx} WHERE category IS NULL AND is_duplicate = 0 ORDER BY date DESC`\n\n const txRes = await sqlite.execute(sql)\n const uncategorized = txRes.rows.map((r: any) => ({\n id: String(r[0]),\n date: String(r[1]),\n description: String(r[2]),\n amount: Number(r[3]),\n abs_amount: Number(r[4]),\n account_name: String(r[5] || ''),\n institution: String(r[6] || ''),\n is_income: Number(r[7]),\n }))\n\n if (uncategorized.length === 0) {\n console.log(`Classification: 0 to classify, ${deduped} deduped`)\n return { classified: 0, skipped: 0, deduped, errors }\n }\n\n const systemPrompt = await getConfig('system_prompt', DEFAULT_SYSTEM_PROMPT)\n const cats = await getCategories()\n let classified = 0\n let skipped = 0\n const batchSize = 50\n const affectedMonths = new Set()\n\n for (let i = 0; i < uncategorized.length; i += batchSize) {\n const batch = uncategorized.slice(i, i + batchSize)\n\n try {\n const { results, error: batchError } = await classifyBatch(apiKey, batch, cats, systemPrompt)\n if (batchError) {\n errors.push(`Batch ${i}: ${batchError}`)\n skipped += batch.length\n continue\n }\n\n for (const r of results) {\n const cat = r.category\n const isTransfer = cat === 'Transfers' || cat === 'Investments' ? 1 : 0\n const tx = batch.find((t) => t.id === r.id)\n const isIncome = cat === 'Income' ? 1 : (tx?.is_income ?? 0)\n try {\n await sqlite.execute({\n sql: `UPDATE ${T.tx} SET category = ?, category_source = 'llm', llm_confidence = ?, llm_reasoning = ?, is_transfer = ?, is_income = ? WHERE id = ? AND (category_source != 'user' OR category_source IS NULL)`,\n args: [cat, r.confidence, r.reasoning, isTransfer, isIncome, r.id],\n })\n classified++\n if (tx) affectedMonths.add(tx.date.slice(0, 7))\n } catch {\n skipped++\n }\n }\n } catch (e: any) {\n errors.push(`Batch ${i}: ${e.message}`)\n skipped += batch.length\n }\n }\n\n for (const month of affectedMonths) await recomputeMonth(month)\n console.log(\n `Classification: ${classified} classified, ${skipped} skipped, ${deduped} deduped, ${errors.length} errors`\n )\n return { classified, skipped, deduped, errors }\n}\n\nexport type PreviewResult = {\n id: string\n date: string\n description: string\n abs_amount: number\n account_name: string\n institution: string\n is_income: number\n category: string\n confidence: number\n reasoning: string\n}\n\nexport async function previewClassification(\n customPrompt?: string\n): Promise<{ results: PreviewResult[]; errors: string[] }> {\n await ensureSchema()\n const apiKey = Deno.env.get('ANTHROPIC_API_KEY')\n if (!apiKey) return { results: [], errors: ['Missing ANTHROPIC_API_KEY'] }\n\n const txRes = await sqlite.execute(\n `SELECT id, date, description, amount, abs_amount, account_name, institution, is_income FROM ${T.tx} WHERE category IS NULL AND is_duplicate = 0 ORDER BY date DESC`\n )\n const uncategorized = txRes.rows.map((r: any) => ({\n id: String(r[0]),\n date: String(r[1]),\n description: String(r[2]),\n amount: Number(r[3]),\n abs_amount: Number(r[4]),\n account_name: String(r[5] || ''),\n institution: String(r[6] || ''),\n is_income: Number(r[7]),\n }))\n\n if (uncategorized.length === 0) return { results: [], errors: [] }\n\n const systemPrompt = customPrompt || (await getConfig('system_prompt', DEFAULT_SYSTEM_PROMPT))\n const cats = await getCategories()\n const allResults: PreviewResult[] = []\n const errors: string[] = []\n const batchSize = 50\n\n for (let i = 0; i < uncategorized.length; i += batchSize) {\n const batch = uncategorized.slice(i, i + batchSize)\n\n try {\n const { results, error: batchError } = await classifyBatch(apiKey, batch, cats, systemPrompt)\n if (batchError) {\n errors.push(`Batch ${i}: ${batchError}`)\n continue\n }\n\n for (const r of results) {\n const tx = batch.find((t) => t.id === r.id)\n if (!tx) continue\n allResults.push({\n ...tx,\n category: r.category,\n confidence: r.confidence,\n reasoning: r.reasoning,\n })\n }\n } catch (e: any) {\n errors.push(`Batch ${i}: ${e.message}`)\n }\n }\n\n return { results: allResults, errors }\n}\n\nexport async function applyClassification(\n results: PreviewResult[]\n): Promise<{ applied: number; errors: string[] }> {\n await ensureSchema()\n const cats = await getCategories()\n let applied = 0\n const errors: string[] = []\n const affectedMonths = new Set()\n\n for (const r of results) {\n const cat = cats.includes(r.category) ? r.category : 'Other'\n const isTransfer = cat === 'Transfers' || cat === 'Investments' ? 1 : 0\n const isIncome = cat === 'Income' ? 1 : r.is_income\n try {\n await sqlite.execute({\n sql: `UPDATE ${T.tx} SET category = ?, category_source = 'llm', llm_confidence = ?, llm_reasoning = ?, is_transfer = ?, is_income = ? WHERE id = ? AND (category_source != 'user' OR category_source IS NULL)`,\n args: [cat, r.confidence, r.reasoning, isTransfer, isIncome, r.id],\n })\n applied++\n affectedMonths.add(r.date.slice(0, 7))\n } catch (e: any) {\n errors.push(`${r.id}: ${e.message}`)\n }\n }\n\n for (const month of affectedMonths) await recomputeMonth(month)\n return { applied, errors }\n}\n\nasync function runDedup(): Promise {\n const res = await sqlite.execute(`\n SELECT a.id, b.id, a.description, b.description, a.institution, b.institution\n FROM ${T.tx} a\n JOIN ${T.tx} b ON a.abs_amount = b.abs_amount\n AND ABS(julianday(a.date) - julianday(b.date)) <= 2\n AND a.id < b.id AND a.institution != b.institution\n WHERE a.is_duplicate = 0 AND b.is_duplicate = 0 AND a.abs_amount > 50\n `)\n\n let deduped = 0\n for (const row of res.rows) {\n const idA = String(row[0]),\n idB = String(row[1])\n const descA = String(row[2]).toUpperCase(),\n descB = String(row[3]).toUpperCase()\n const genericPatterns = ['VENMO PAYMENT', 'ZELLE PAYMENT', 'PAYPAL']\n const aIsGeneric = genericPatterns.some((p) => descA.includes(p))\n const bIsGeneric = genericPatterns.some((p) => descB.includes(p))\n\n if (aIsGeneric && !bIsGeneric) {\n await sqlite.execute({\n sql: `UPDATE ${T.tx} SET is_duplicate = 1, duplicate_of = ?, is_transfer = 1, category = 'Transfers', category_source = 'dedup' WHERE id = ?`,\n args: [idB, idA],\n })\n deduped++\n } else if (bIsGeneric && !aIsGeneric) {\n await sqlite.execute({\n sql: `UPDATE ${T.tx} SET is_duplicate = 1, duplicate_of = ?, is_transfer = 1, category = 'Transfers', category_source = 'dedup' WHERE id = ?`,\n args: [idA, idB],\n })\n deduped++\n }\n }\n return deduped\n}\n",[176],"2c7ab59e-5ad0-11f1-9b25-ee650bb23af1","app.ts",202,["D",1779998557094],26,"http","// Spend Agent — Lean burn tracker\n// Home: stacked bar chart + burn number + category breakdown (expandable)\n// Settings: connect bank + classification prompt + CSV upload\nimport {\n sqlite,\n ensureSchema,\n getConfig,\n setConfig,\n DEFAULT_SYSTEM_PROMPT,\n getCategories,\n saveCategoriesAndCleanup,\n loadCategoryColors,\n recomputeMonth,\n seedDemoData,\n insertSimulatedTransactions,\n T,\n} from './db.ts'\nimport { runClassification, previewClassification, applyClassification } from './classify-engine.ts'\nimport { classifyDrawerHtml } from './classify-ui.ts'\nimport { runSync } from './sync-engine.ts'\nimport { blob } from 'https://esm.town/v/std/blob'\nimport {\n fmt,\n escHtml,\n CAT_COLORS,\n pageShell,\n TELLER_APP_ID,\n CATEGORIES_JS,\n CATEGORY_MODAL_HTML,\n} from './lib.ts'\n\nexport default async function (req: Request): Promise {\n await ensureSchema()\n await seedDemoData()\n await loadCategoryColors()\n const url = new URL(req.url)\n const path = url.pathname\n\n if (path === '/sync' && req.method === 'POST') return handleSync()\n if (path === '/classify' && req.method === 'POST') return handleClassify(false)\n if (path === '/reclassify' && req.method === 'POST') return handleClassify(true)\n if (path === '/save-token' && req.method === 'POST') return handleSaveToken(req)\n if (path === '/update-category' && req.method === 'POST') return handleUpdateCategory(req)\n if (path === '/save-prompt' && req.method === 'POST') return handleSavePrompt(req)\n if (path === '/save-categories' && req.method === 'POST') return handleSaveCategories(req)\n if (path === '/upload-csv' && req.method === 'POST') return handleCsvUpload(req)\n if (path === '/category-count') return handleCategoryCount(url)\n if (path === '/simulate-transactions' && req.method === 'POST')\n return handleSimulateTransactions()\n\n if (path === '/preview-classify' && req.method === 'POST') return handlePreviewClassify(req)\n if (path === '/approve-classify' && req.method === 'POST') return handleApproveClassify(req)\n if (path === '/add-rule' && req.method === 'POST') return handleAddRule(req)\n if (path === '/ask' && req.method === 'POST') return handleAsk(req)\n if (path === '/diagram.png') {\n const png = await blob.get('spend-agent-diagram')\n return new Response(png.body, {\n headers: { 'Content-Type': 'image/png', 'Cache-Control': 'public, max-age=86400' },\n })\n }\n if (path === '/ask') return html(await askPage())\n if (path === '/settings') return html(await settingsPage())\n return html(await burnPage(url))\n}\n\nfunction html(body: string): Response {\n return new Response(body, { headers: { 'Content-Type': 'text/html' } })\n}\n\nasync function handleSimulateTransactions(): Promise {\n try {\n const inserted = await insertSimulatedTransactions()\n return Response.json({ inserted })\n } catch (e: any) {\n return Response.json({ error: e.message }, { status: 500 })\n }\n}\n\nasync function handleCategoryCount(url: URL): Promise {\n const name = url.searchParams.get('name') || ''\n if (!name) return Response.json({ count: 0 })\n const res = await sqlite.execute({\n sql: `SELECT COUNT(*) as cnt FROM ${T.tx} WHERE category = ?`,\n args: [name],\n })\n const count = Number((res.rows[0] as any)[0]) || 0\n return Response.json({ count })\n}\n\n// ── Burn Page (Home) ────────────────────────────────────────────\nasync function burnPage(url: URL): Promise {\n const now = new Date()\n const monthParam = url.searchParams.get('month') || ''\n const curMonth =\n monthParam || `${now.getFullYear()}-${String(now.getMonth() + 1).padStart(2, '0')}`\n\n // Check setup state for empty-state\n const hasCert = !!Deno.env.get('TELLER_CERT')\n const tokensRes = await sqlite.execute('SELECT COUNT(*) FROM teller_tokens')\n const hasAccounts = Number(tokensRes.rows[0]?.[0]) > 0\n const txCountRes = await sqlite.execute('SELECT COUNT(*) FROM transactions')\n const txTotal = Number(txCountRes.rows[0]?.[0]) || 0\n\n if (txTotal === 0) {\n const steps = []\n if (!TELLER_APP_ID) {\n steps.push(`

Get started

Upload a CSV below, or connect a bank by setting TELLER_APP_ID in your env vars. Sandbox works without certificates — add TELLER_CERT + TELLER_KEY for real banks.

`)\n } else if (!hasAccounts) {\n steps.push(`

Connect your bank

Link a checking or credit card account to start tracking.

\n Go to Settings\n

`)\n } else {\n steps.push(`

Syncing transactions...

Your bank is connected. Transactions will appear after the next sync.

\n \n

`)\n }\n steps.push(`

Or upload a CSV export from your credit card or Venmo

\n Upload CSV\n

`)\n const scripts = ``\n return pageShell('Burn', '/', steps.join(''), scripts)\n }\n\n // ── Stacked bar chart data: per-month per-category ──\n const stackRes = await sqlite.execute(\n `SELECT substr(date,1,7) AS m, COALESCE(category, 'Unclassified') AS cat, SUM(abs_amount) AS spend\n FROM transactions\n WHERE is_income = 0 AND is_transfer = 0 AND is_duplicate = 0\n GROUP BY m, cat\n ORDER BY m DESC`\n )\n\n // Build per-month structure\n const monthMap: Record<\n string,\n { month: string; total: number; cats: { cat: string; spend: number }[] }\n > = {}\n for (const r of stackRes.rows) {\n const m = String(r[0])\n const cat = String(r[1])\n const spend = Number(r[2])\n if (!monthMap[m]) monthMap[m] = { month: m, total: 0, cats: [] }\n monthMap[m].total += spend\n monthMap[m].cats.push({ cat, spend })\n }\n\n const allMonths = Object.values(monthMap)\n .sort((a, b) => a.month.localeCompare(b.month))\n .slice(-12)\n const maxSpend = Math.max(...allMonths.map((m) => m.total), 1)\n\n // Per-month income for net chart\n const incomeStackRes = await sqlite.execute(\n `SELECT substr(date,1,7) AS m, SUM(abs_amount) AS inc\n FROM transactions\n WHERE is_income = 1 AND is_transfer = 0 AND is_duplicate = 0\n GROUP BY m ORDER BY m DESC`\n )\n const monthIncomeMap: Record = {}\n for (const r of incomeStackRes.rows) {\n monthIncomeMap[String(r[0])] = Number(r[1])\n }\n\n // Current month spend\n const curData = allMonths.find((m) => m.month === curMonth)\n const monthSpend = curData?.total || 0\n\n // Unclassified count\n const uncRes = await sqlite.execute({\n sql: 'SELECT COUNT(*) FROM transactions WHERE category IS NULL AND is_duplicate = 0 AND substr(date,1,7) = ?',\n args: [curMonth],\n })\n const uncCount = Number(uncRes.rows[0]?.[0]) || 0\n\n // Category breakdown for current month\n const cats = await getCategories()\n const catRes = await sqlite.execute({\n sql: `SELECT category, SUM(abs_amount) as total, COUNT(*) as cnt\n FROM transactions WHERE substr(date,1,7) = ? AND is_income = 0 AND is_transfer = 0 AND is_duplicate = 0\n GROUP BY category ORDER BY SUM(abs_amount) DESC`,\n args: [curMonth],\n })\n\n // Income total for current month\n const incRes = await sqlite.execute({\n sql: `SELECT SUM(abs_amount) FROM transactions\n WHERE substr(date,1,7) = ? AND is_income = 1 AND is_transfer = 0 AND is_duplicate = 0`,\n args: [curMonth],\n })\n const monthIncome = Number(incRes.rows[0]?.[0]) || 0\n const netCashFlow = monthIncome - monthSpend\n\n // All transactions for this month\n const txRes = await sqlite.execute({\n sql: `SELECT id, date, description, abs_amount, category, category_source, llm_confidence, llm_reasoning, is_income, is_transfer, account_name, institution\n FROM transactions WHERE substr(date,1,7) = ? AND is_duplicate = 0\n ORDER BY date DESC, abs_amount DESC`,\n args: [curMonth],\n })\n\n // Group by category + collect income transactions\n const txByCategory: Record = {}\n const incomeTxs: any[] = []\n for (const r of txRes.rows) {\n const cat = r[4] ? String(r[4]) : 'Unclassified'\n const isIncome = Number(r[8])\n const isTransfer = Number(r[9])\n const txObj = {\n id: String(r[0]),\n date: String(r[1]),\n desc: String(r[2]),\n amount: Number(r[3]),\n category: r[4] ? String(r[4]) : null,\n source: r[5] ? String(r[5]) : null,\n confidence: r[6] != null ? Number(r[6]) : null,\n reasoning: r[7] ? String(r[7]) : null,\n account: String(r[10] || ''),\n institution: String(r[11] || ''),\n }\n if (isTransfer) continue\n if (isIncome) {\n incomeTxs.push(txObj)\n } else {\n if (!txByCategory[cat]) txByCategory[cat] = []\n txByCategory[cat].push(txObj)\n }\n }\n\n // ── Stacked bar chart HTML ──\n // Collect global category totals for legend (top 7)\n const globalCatTotals: Record = {}\n for (const m of allMonths) {\n for (const c of m.cats) {\n globalCatTotals[c.cat] = (globalCatTotals[c.cat] || 0) + c.spend\n }\n }\n const legendCats = Object.entries(globalCatTotals)\n .sort((a, b) => b[1] - a[1])\n .slice(0, 7)\n .map(([cat]) => cat)\n\n const barChart = allMonths\n .map((m) => {\n const totalH = Math.max(Math.round((m.total / maxSpend) * 120), 2)\n const active = m.month === curMonth\n const label = new Date(m.month + '-15').toLocaleDateString('en-US', {\n month: 'short',\n })\n const shortYear = m.month.slice(2, 4)\n\n // Consolidate small categories (<4% of month) into \"Other\" for cleaner bars\n const threshold = m.total * 0.04\n const bigCats: { cat: string; spend: number }[] = []\n let smallSum = 0\n for (const c of m.cats) {\n if (c.spend >= threshold) bigCats.push(c)\n else smallSum += c.spend\n }\n if (smallSum > 0) bigCats.push({ cat: 'Other', spend: smallSum })\n bigCats.sort((a, b) => a.spend - b.spend)\n\n const segments = bigCats\n .map((c) => {\n const segH = Math.max(Math.round((c.spend / m.total) * totalH), 2)\n const color = CAT_COLORS[c.cat] || '#6b7280'\n return `

`\n })\n .join('')\n\n return `

${m.total >= 1000 ? '$' + (m.total / 1000).toFixed(1) + 'k' : '$' + Math.round(m.total)}

${segments}

${label}
'${shortYear}

`\n })\n .join('')\n\n const legend = legendCats\n .map((cat) => {\n const color = CAT_COLORS[cat] || '#6b7280'\n return `${cat}`\n })\n .join('')\n\n // ── Net income bar chart ──\n const netMonths = allMonths.map((m) => {\n const inc = monthIncomeMap[m.month] || 0\n const net = inc - m.total\n return { month: m.month, spend: m.total, income: inc, net }\n })\n const maxNetAbs = Math.max(...netMonths.map((m) => Math.abs(m.net)), 1)\n\n const netBarChart = netMonths\n .map((m) => {\n const active = m.month === curMonth\n const label = new Date(m.month + '-15').toLocaleDateString('en-US', { month: 'short' })\n const shortYear = m.month.slice(2, 4)\n const barH = Math.max(Math.round((Math.abs(m.net) / maxNetAbs) * 100), 2)\n const color = m.net >= 0 ? '#22c55e' : '#ef4444'\n const sign = m.net >= 0 ? '+' : '-'\n const valStr =\n Math.abs(m.net) >= 1000\n ? `${sign}$${(Math.abs(m.net) / 1000).toFixed(1)}k`\n : `${sign}$${Math.round(Math.abs(m.net))}`\n\n return `

${valStr}

${label}
'${shortYear}

`\n })\n .join('')\n\n const netLegend = `Positive\n Negative`\n\n // Category accordion\n const catRows = catRes.rows\n .map((r: any) => {\n const cat = String(r[0] || 'Unclassified')\n const total = Number(r[1])\n const count = Number(r[2])\n const color = CAT_COLORS[cat] || '#6b7280'\n const pctVal = monthSpend > 0 ? Math.round((total / monthSpend) * 100) : 0\n const barW = monthSpend > 0 ? Math.round((total / monthSpend) * 100) : 0\n const slug = cat.toLowerCase().replace(/[^a-z]/g, '')\n\n const txs = txByCategory[cat] || []\n const txRowsHtml = txs\n .map((tx) => {\n const dateLabel = new Date(tx.date + 'T12:00:00').toLocaleDateString('en-US', {\n month: 'short',\n day: 'numeric',\n })\n const confBadge =\n tx.confidence != null && tx.confidence < 0.7\n ? 'LOW'\n : ''\n const catOptions = cats\n .map(\n (c: string) =>\n ``\n )\n .join('')\n\n return `

\n ${dateLabel}\n ${escHtml(tx.desc.slice(0, 40))}\n -${fmt(tx.amount)}\n

${escHtml(tx.desc)}

\n ${escHtml(tx.account)}${tx.institution ? ' · ' + escHtml(tx.institution) : ''}\n ${tx.reasoning ? `${escHtml(tx.reasoning)}` : ''}\n

\n ${confBadge}\n

`\n })\n .join('')\n\n return `

\n \n ${escHtml(cat)}\n ${count}\n

\n ${fmt(total)}\n ${pctVal}%\n ▶\n

${txRowsHtml}

`\n })\n .join('')\n\n // Collect unclassified transactions for drawer\n const uncTxs = (txByCategory['Unclassified'] || []).map((tx: any) => ({\n id: tx.id,\n date: tx.date,\n description: tx.desc,\n abs_amount: tx.amount,\n account_name: tx.account,\n institution: tx.institution,\n }))\n const currentPrompt = await getConfig('system_prompt', DEFAULT_SYSTEM_PROMPT)\n const drawerUi = classifyDrawerHtml(uncCount, uncTxs, currentPrompt)\n const uncBanner = drawerUi.banner\n\n const monthLabel = new Date(curMonth + '-15').toLocaleDateString('en-US', {\n month: 'long',\n year: 'numeric',\n })\n\n // Group income by category\n const incomeByCategory: Record = {}\n for (const tx of incomeTxs) {\n const cat = tx.category || 'Income'\n if (!incomeByCategory[cat]) incomeByCategory[cat] = { txs: [], total: 0 }\n incomeByCategory[cat].txs.push(tx)\n incomeByCategory[cat].total += tx.amount\n }\n // ── Budget view: merged spend + income per category ──\n const spendByCategory: Record = {}\n for (const r of catRes.rows) {\n spendByCategory[String(r[0] || 'Unclassified')] = Number(r[1])\n }\n const allBudgetCats = new Set([...Object.keys(spendByCategory), ...Object.keys(incomeByCategory)])\n const budgetRows = [...allBudgetCats]\n .map((cat) => {\n const spend = spendByCategory[cat] || 0\n const income = incomeByCategory[cat]?.total || 0\n const net = income - spend\n const spendTxs = txByCategory[cat] || []\n const incTxs = incomeByCategory[cat]?.txs || []\n return { cat, spend, income, net, spendTxs, incTxs, totalActivity: spend + income }\n })\n .sort((a, b) => b.totalActivity - a.totalActivity)\n\n const budgetCatRows = budgetRows\n .map((row) => {\n const color = CAT_COLORS[row.cat] || '#6b7280'\n const slug = 'bgt-' + row.cat.toLowerCase().replace(/[^a-z]/g, '')\n const netColor = row.net >= 0 ? '#22c55e' : '#ef4444'\n const netSign = row.net >= 0 ? '+' : '-'\n const txCount = row.spendTxs.length + row.incTxs.length\n // Sub-labels showing spend and income breakdown\n const parts: string[] = []\n if (row.spend > 0) parts.push(`-${fmt(row.spend)}`)\n if (row.income > 0) parts.push(`+${fmt(row.income)}`)\n const breakdown =\n parts.length === 2\n ? `(${parts.join(' ')})`\n : ''\n\n // Mix spend + income transactions sorted by date\n const allTxs = [\n ...row.spendTxs.map((tx: any) => ({ ...tx, _type: 'spend' })),\n ...row.incTxs.map((tx: any) => ({ ...tx, _type: 'income' })),\n ].sort((a: any, b: any) => b.date.localeCompare(a.date))\n\n const txRowsHtml = allTxs\n .map((tx: any) => {\n const dateLabel = new Date(tx.date + 'T12:00:00').toLocaleDateString('en-US', {\n month: 'short',\n day: 'numeric',\n })\n const amtColor = tx._type === 'income' ? '#22c55e' : '#ef4444'\n const amtSign = tx._type === 'income' ? '+' : '-'\n const catOptions = cats\n .map(\n (c: string) =>\n ``\n )\n .join('')\n return `

\n ${dateLabel}\n ${escHtml(tx.desc.slice(0, 40))}\n ${amtSign}${fmt(tx.amount)}\n

${escHtml(tx.desc)}

\n ${escHtml(tx.account)}${tx.institution ? ' · ' + escHtml(tx.institution) : ''}\n ${tx.reasoning ? `${escHtml(tx.reasoning)}` : ''}\n

\n \n

`\n })\n .join('')\n\n return `

\n \n ${escHtml(row.cat)}\n ${txCount}\n

\n ${netSign}${fmt(Math.abs(row.net))}\n ${breakdown}\n ▶\n

${txRowsHtml}

`\n })\n .join('')\n\n const body = `\n ${uncBanner}\n

${barChart}

${legend}

${netBarChart}

${netLegend}

\n ${monthLabel}\n

\n \n \n

Total Spend

\n ${fmt(monthSpend)}\n

Income

\n ${monthIncome > 0 ? '+' : ''}${fmt(monthIncome)}\n

−

Spend

\n ${fmt(monthSpend)}\n

Net

\n = 0 ? '#22c55e' : '#ef4444'}\">${netCashFlow >= 0 ? '+' : '-'}${fmt(Math.abs(netCashFlow))}\n

${catRows || '

No spending this month

\n \n ${drawerUi.drawer}`\n\n const scripts = ``\n\n return pageShell('Burn', '/', body, scripts)\n}\n\n// ── Ask Page (Agent Q&A) ────────────────────────────────────────\nasync function askPage(): Promise {\n const suggestions = [\n \"Where's my money going this month?\",\n 'How does this month compare to last?',\n 'Any new subscriptions I should know about?',\n \"What's my biggest spending spike?\",\n 'Am I spending more on food than usual?',\n 'What can I cut to save $500/month?',\n ]\n\n const chips = suggestions\n .map(\n (s) => ``\n )\n .join('')\n\n const body = `\n

Ask about your spending

AI-powered insights from your real transaction data

${chips}

\n \n

\n \n \n \n \n

`\n\n const scripts = ``\n\n return pageShell('Ask', '/ask', body, scripts)\n}\n\nasync function gatherFinancialContext(): Promise {\n const now = new Date()\n const curMonth = `${now.getFullYear()}-${String(now.getMonth() + 1).padStart(2, '0')}`\n\n // Last 6 months summaries\n const summaries = await sqlite.execute(\n `SELECT month, total_spend, total_income, category_breakdown, tx_count\n FROM monthly_summaries ORDER BY month DESC LIMIT 6`\n )\n\n let ctx = `## Monthly Summaries (last 6 months)\\n`\n for (const r of summaries.rows) {\n const month = String(r[0])\n const spend = Number(r[1])\n const income = Number(r[2])\n const cats = r[3] ? String(r[3]) : '{}'\n ctx += `${month}: $${spend.toFixed(0)} spent, $${income.toFixed(0)} income | ${cats}\\n`\n }\n\n // Current month transactions (last 50)\n const txRes = await sqlite.execute({\n sql: `SELECT date, description, abs_amount, category, account_name, institution, is_income, is_transfer\n FROM transactions\n WHERE substr(date,1,7) = ? AND is_duplicate = 0\n ORDER BY date DESC LIMIT 50`,\n args: [curMonth],\n })\n\n ctx += `\\n## Current month transactions (${curMonth}, most recent 50)\\n`\n ctx += `date | description | amount | category | account\\n`\n for (const r of txRes.rows) {\n const isIncome = Number(r[6])\n const isTransfer = Number(r[7])\n const tag = isIncome ? ' [INCOME]' : isTransfer ? ' [TRANSFER]' : ''\n ctx += `${r[0]} | ${String(r[1]).slice(0, 50)} | $${Number(r[2]).toFixed(2)}${tag} | ${r[3] || 'Unclassified'} | ${r[4] || ''}\\n`\n }\n\n // Recurring charges detection\n const recurRes = await sqlite.execute(\n `SELECT description, COUNT(*) as cnt, ROUND(AVG(abs_amount),2) as avg_amt\n FROM transactions\n WHERE is_income = 0 AND is_transfer = 0 AND is_duplicate = 0\n GROUP BY LOWER(TRIM(description))\n HAVING cnt >= 3\n ORDER BY cnt DESC LIMIT 20`\n )\n\n ctx += `\\n## Likely recurring charges (appeared 3+ times)\\n`\n for (const r of recurRes.rows) {\n ctx += `${r[0]} — ${r[1]}x, avg $${r[2]}\\n`\n }\n\n // Recent user corrections (for context)\n const overrides = await sqlite.execute(\n `SELECT description, category FROM transactions\n WHERE category_source = 'user' ORDER BY created_at DESC LIMIT 20`\n )\n if (overrides.rows.length > 0) {\n ctx += `\\n## Recent user corrections (these override LLM classifications)\\n`\n for (const r of overrides.rows) {\n ctx += `${String(r[0]).slice(0, 50)} → ${r[1]}\\n`\n }\n }\n\n return ctx\n}\n\nasync function handleAsk(req: Request): Promise {\n try {\n const { question, images } = await req.json()\n if (!question && (!images || images.length === 0))\n return Response.json({ error: 'No question' }, { status: 400 })\n\n const apiKey = Deno.env.get('ANTHROPIC_API_KEY')\n if (!apiKey)\n return Response.json({\n answer: 'ANTHROPIC_API_KEY not set. Add it in Val Town environment variables.',\n })\n\n const context = await gatherFinancialContext()\n const hasImages = images && images.length > 0\n\n // Build multimodal content blocks\n const content: any[] = []\n if (hasImages) {\n for (const img of images) {\n const match = img.match(/^data:(image\\/[a-zA-Z+]+);base64,(.+)$/)\n if (match) {\n content.push({\n type: 'image',\n source: { type: 'base64', media_type: match[1], data: match[2] },\n })\n }\n }\n }\n content.push({\n type: 'text',\n text:\n question || 'What do you see in this screenshot? Suggest any classification corrections.',\n })\n\n const systemPrompt = `You are a helpful personal finance assistant analyzing the user's real transaction data.${hasImages ? ' The user has shared a screenshot — analyze it alongside their transaction data to identify any misclassified transactions.' : ''} Be concise, specific, and use actual dollar amounts from the data. Format responses with markdown: use **bold** for key numbers and totals, bullet lists for category breakdowns, and ### headings to organize sections. Keep paragraphs short. If you spot concerning patterns, flag them directly. Don't be preachy — be useful.${hasImages ? '\\n\\nWhen suggesting category corrections, list each one clearly: the transaction description, its current category, and what it should be changed to.' : ''}\\n\\n${context}`\n\n const res = await fetch('https://api.anthropic.com/v1/messages', {\n method: 'POST',\n headers: {\n 'x-api-key': apiKey,\n 'anthropic-version': '2023-06-01',\n 'content-type': 'application/json',\n },\n body: JSON.stringify({\n model: 'claude-haiku-4-5-20251001',\n max_tokens: 2048,\n system: systemPrompt,\n messages: [{ role: 'user', content }],\n }),\n })\n\n const data = await res.json()\n const answer = data.content?.[0]?.text || 'No response from AI.'\n return Response.json({ answer })\n } catch (e: any) {\n return Response.json({ answer: 'Error: ' + e.message })\n }\n}\n\n// ── Settings Page ───────────────────────────────────────────────\nasync function settingsPage(): Promise {\n const prompt = await getConfig('system_prompt', DEFAULT_SYSTEM_PROMPT)\n const tellerEnv = Deno.env.get('TELLER_ENV') || 'sandbox'\n const isSandbox = tellerEnv === 'sandbox'\n const categories = await getCategories()\n const categoriesListHtml = categories\n .map((c) => {\n const o = c === 'Other'\n const s = escHtml(c)\n const col = CAT_COLORS[c] || '#6b7280'\n return `

${o ? 'required' : ''}

`\n })\n .join('')\n\n // Teller tokens\n const tokensRes = await sqlite.execute(\n 'SELECT id, institution, linked_at FROM teller_tokens ORDER BY linked_at DESC'\n )\n const tellerTokens = tokensRes.rows.map((r: any) => ({\n id: String(r[0]),\n inst: String(r[1]),\n linked: String(r[2]),\n }))\n\n // Sandbox credentials disclosed via a small

accordion below the\n // \"Get one free\" / \"Connect Bank\" affordance. Discoverable but unobtrusive.\n // Source: https://teller.io/docs/guides/sandbox\n const sandboxCredsContent = isSandbox\n ? `\n `\n : ''\n\n function renderAccountList(\n tokens: { id: string; inst: string; linked: string }[],\n source: string\n ) {\n if (tokens.length === 0)\n return `No ${source} accounts linked`\n return tokens\n .map(\n (\n t\n ) => `

\n \n ${escHtml(t.inst)}\n Linked ${new Date(t.linked).toLocaleDateString('en-US', { month: 'short', day: 'numeric' })}\n \n

`\n )\n .join('')\n }\n\n const body = `\n

Teller Accounts

\n Banks & credit cards\n

\n ${renderAccountList(tellerTokens, 'Teller')}\n ${!TELLER_APP_ID ? `

The accounts above are demo placeholders. Set TELLER_APP_ID in env vars to connect a real (sandbox or production) bank. Get one free at teller.io

` : ''}\n ${TELLER_APP_ID ? `

` : ''}\n

\n ${isSandbox ? `` : ''}\n \n

\n ${sandboxCredsContent}\n

CSV Import

\n \n \n

\n \n

Select a format above first

\n \n

\n \n \n \n

Classification Prompt

Tell the AI how to categorize. Use natural language.

\n \n

\n \n \n

Categories

Feeds the AI classifier's allowed set (tool-use enum). Removing one sends its transactions back to unclassified for re-review. Other is required.

\n \n \n

\n ${CATEGORY_MODAL_HTML}`\n\n const scripts = `${TELLER_APP_ID ? '' : ''}\n `\n\n return pageShell('Settings', '/settings', body, scripts)\n}\n\n// ── API Handlers ────────────────────────────────────────────────\nasync function handleSync(): Promise {\n try {\n return Response.json(await runSync())\n } catch (e: any) {\n return Response.json({ error: e.message }, { status: 500 })\n }\n}\n\nasync function handleClassify(reclassifyAll: boolean): Promise {\n try {\n return Response.json(await runClassification(reclassifyAll))\n } catch (e: any) {\n return Response.json({ error: e.message }, { status: 500 })\n }\n}\n\nasync function handlePreviewClassify(req: Request): Promise {\n try {\n const body = await req.json().catch(() => ({}))\n const result = await previewClassification(body.prompt || undefined)\n return Response.json(result)\n } catch (e: any) {\n return Response.json({ error: e.message }, { status: 500 })\n }\n}\n\nasync function handleApproveClassify(req: Request): Promise {\n try {\n const { results, prompt } = await req.json()\n if (prompt) await setConfig('system_prompt', prompt)\n const result = await applyClassification(results)\n return Response.json(result)\n } catch (e: any) {\n return Response.json({ error: e.message }, { status: 500 })\n }\n}\n\nasync function handleSaveToken(req: Request): Promise {\n const { accessToken, enrollment } = await req.json()\n const instName = enrollment?.institution?.name || 'Unknown'\n const existing = await sqlite.execute({\n sql: 'SELECT id FROM teller_tokens WHERE institution = ?',\n args: [instName],\n })\n if (existing.rows.length > 0) {\n await sqlite.execute({\n sql: \"UPDATE teller_tokens SET access_token = ?, enrollment_id = ?, linked_at = datetime('now') WHERE institution = ?\",\n args: [accessToken, enrollment?.id, instName],\n })\n } else {\n const id = crypto.randomUUID()\n await sqlite.execute({\n sql: 'INSERT INTO teller_tokens (id, access_token, enrollment_id, institution) VALUES (?, ?, ?, ?)',\n args: [id, accessToken, enrollment?.id, instName],\n })\n }\n let syncResult = {\n accounts: 0,\n inserted: 0,\n skipped: 0,\n errors: [] as string[],\n }\n try {\n syncResult = await runSync()\n } catch {}\n if (syncResult.inserted > 0) {\n try {\n await runClassification()\n } catch {}\n }\n return Response.json({ ok: true, institution: instName, ...syncResult })\n}\n\nasync function handleUpdateCategory(req: Request): Promise {\n const { id, category } = await req.json()\n\n // Get full transaction context before updating (for intelligent learned rule)\n const txRes = await sqlite.execute({\n sql: `SELECT description, category, date, abs_amount, source, account_name, institution\n FROM transactions WHERE id = ?`,\n args: [id],\n })\n if (txRes.rows.length === 0) return Response.json({ ok: false, error: 'Transaction not found' })\n\n const row = txRes.rows[0]\n const oldCategory = String(row[1] || 'Unclassified')\n const txContext = {\n description: String(row[0]),\n date: String(row[2]),\n amount: Number(row[3]),\n source: String(row[4] || ''),\n account: String(row[5] || ''),\n institution: String(row[6] || ''),\n }\n\n await sqlite.execute({\n sql: \"UPDATE transactions SET category = ?, category_source = 'user', llm_confidence = NULL, llm_reasoning = NULL WHERE id = ?\",\n args: [category || null, id],\n })\n\n const month = txContext.date.slice(0, 7)\n await recomputeMonth(month)\n\n // Generate intelligent learned rule with full context\n if (category && category !== oldCategory) {\n await appendLearnedRule(txContext, category, oldCategory)\n }\n\n return Response.json({ ok: true })\n}\n\nconst LEARNED_RULES_HEADER = '\\n\\n## Learned from your corrections'\n\nasync function generateSmartRule(\n tx: {\n description: string\n date: string\n amount: number\n source: string\n account: string\n institution: string\n },\n newCategory: string,\n oldCategory: string\n): Promise {\n const apiKey = Deno.env.get('ANTHROPIC_API_KEY')\n if (!apiKey) {\n // Fallback to template rule if no API key\n return `${tx.description.trim().slice(0, 50)} (${tx.institution || tx.source}, ~$${Math.round(tx.amount)}) → ${newCategory}`\n }\n\n try {\n const res = await fetch('https://api.anthropic.com/v1/messages', {\n method: 'POST',\n headers: {\n 'Content-Type': 'application/json',\n 'x-api-key': apiKey,\n 'anthropic-version': '2023-06-01',\n },\n body: JSON.stringify({\n model: 'claude-haiku-4-5-20251001',\n max_tokens: 150,\n messages: [\n {\n role: 'user',\n content: `Write a ONE-LINE classification rule for a personal finance categorizer. The user corrected this transaction from \"${oldCategory}\" to \"${newCategory}\".\n\nTransaction: \"${tx.description}\"\nAmount: $${tx.amount.toFixed(2)}\nSource: ${tx.source} (${tx.institution}, ${tx.account})\nDate: ${tx.date}\n\nWrite a concise rule that captures the PATTERN (not just this one transaction). Include relevant context like the payment platform, merchant name, amount range, or frequency if it helps. Format: \"pattern description → ${newCategory}\"\n\nExamples of good rules:\n- Venmo payments to \"May\" (~$3,300, monthly) → Housing (rent via Venmo)\n- DoorDash orders from San Francisco → Food\n- SCV Labs LLC ACH payments (~$2,000-5,000) → Transfers (own business)\n- Monthly Spotify charge (~$16) → Subscriptions\n\nRespond with ONLY the rule text, nothing else.`,\n },\n ],\n }),\n })\n\n if (res.ok) {\n const data = await res.json()\n const ruleText = data.content?.[0]?.text?.trim()\n if (ruleText && ruleText.length < 200) return ruleText\n }\n } catch {}\n\n // Fallback: enriched template\n const src = tx.institution || tx.source || 'unknown'\n return `${tx.description.trim().slice(0, 50)} (${src}, ~$${Math.round(tx.amount)}) → ${newCategory}`\n}\n\nasync function appendLearnedRule(\n tx: {\n description: string\n date: string\n amount: number\n source: string\n account: string\n institution: string\n },\n newCategory: string,\n oldCategory: string\n) {\n const prompt = await getConfig('system_prompt', DEFAULT_SYSTEM_PROMPT)\n const cleanDesc = tx.description.trim().slice(0, 50)\n\n // Generate an intelligent rule\n const ruleText = await generateSmartRule(tx, newCategory, oldCategory)\n const newRule = `\\n- ${ruleText}`\n\n // Check if rule already exists for this description pattern\n if (prompt.includes(cleanDesc)) {\n // Find and replace the existing rule line containing this description\n const lines = prompt.split('\\n')\n const updatedLines = lines.map((line) => {\n if (line.includes(cleanDesc) && line.trimStart().startsWith('- ')) {\n return `- ${ruleText}`\n }\n return line\n })\n await setConfig('system_prompt', updatedLines.join('\\n'))\n return\n }\n\n // Append new rule\n if (prompt.includes(LEARNED_RULES_HEADER)) {\n await setConfig('system_prompt', prompt + newRule)\n } else {\n await setConfig('system_prompt', prompt + LEARNED_RULES_HEADER + newRule)\n }\n}\n\nasync function handleSavePrompt(req: Request): Promise {\n const { prompt } = await req.json()\n await setConfig('system_prompt', prompt)\n return Response.json({ ok: true })\n}\n\nasync function handleSaveCategories(req: Request): Promise {\n try {\n const body = await req.json()\n let rows = body?.rows\n // Legacy clients (string list) still work: treat each as kept-as-is, no color.\n if (!rows && Array.isArray(body?.categories)) {\n rows = body.categories.map((c: string) => ({ original: c, current: c }))\n }\n if (!Array.isArray(rows)) {\n return Response.json({ ok: false, error: 'rows must be an array' }, { status: 400 })\n }\n const result = await saveCategoriesAndCleanup(rows)\n return Response.json({ ok: true, ...result })\n } catch (e: any) {\n return Response.json({ ok: false, error: e?.message || String(e) }, { status: 500 })\n }\n}\n\nasync function handleAddRule(req: Request): Promise {\n try {\n const { rule } = await req.json()\n if (!rule) return Response.json({ error: 'No rule' }, { status: 400 })\n const prompt = await getConfig('system_prompt', DEFAULT_SYSTEM_PROMPT)\n const newRule = `\\n- ${rule.trim()}`\n if (prompt.includes(LEARNED_RULES_HEADER)) {\n await setConfig('system_prompt', prompt + newRule)\n } else {\n await setConfig('system_prompt', prompt + LEARNED_RULES_HEADER + newRule)\n }\n return Response.json({ ok: true })\n } catch (e: any) {\n return Response.json({ error: e.message }, { status: 500 })\n }\n}\n\nasync function handleCsvUpload(req: Request): Promise {\n try {\n const form = await req.formData()\n const file = form.get('file') as File\n const type = form.get('type') as string\n if (!file) return Response.json({ ok: false, error: 'No file' })\n const text = await file.text()\n const lines = text.split('\\n')\n let inserted = 0,\n skipped = 0\n const affectedMonths = new Set()\n const rows: { date: string; desc: string; amount: number }[] = []\n\n if (type === 'credit_card') {\n // Generic credit card CSV parser — auto-detects column layout from headers\n const headerLine = parseCSVLine(lines[0]).map((h) => h.toLowerCase().trim())\n const dateCol = headerLine.findIndex(\n (h) => /^(transaction\\s*)?date$/.test(h) || h === 'trans date' || h === 'posting date'\n )\n const descCol = headerLine.findIndex(\n (h) =>\n /^desc(ription)?$/.test(h) ||\n h === 'merchant' ||\n h === 'merchant name' ||\n h === 'payee' ||\n h === 'name'\n )\n const amountCol = (() => {\n // Prefer 'amount', fall back to 'amt', 'charge', 'debit'\n const i = headerLine.findLastIndex((h) => /^(amount|amt|charge|debit)$/.test(h))\n return i >= 0\n ? i\n : headerLine.findLastIndex((h) => h.includes('amount') || h.includes('amt'))\n })()\n if (dateCol < 0 || descCol < 0 || amountCol < 0)\n return Response.json({\n ok: false,\n error:\n 'Could not find Date, Description, and Amount columns in header. Found: ' +\n headerLine.join(', '),\n })\n // Sign convention: Chase amounts are negative for charges, Amex are positive\n const isChaseStyle = headerLine.some((h) => h === 'post date' || h === 'posting date')\n // Detect institution from headers\n const institution = isChaseStyle ? 'Chase' : 'American Express'\n const accountName = isChaseStyle ? 'Chase Card' : 'Amex Card'\n const sourceTag = isChaseStyle ? 'csv_chase' : 'csv_amex'\n\n for (let i = 1; i < lines.length; i++) {\n const parts = parseCSVLine(lines[i])\n if (parts.length <= Math.max(dateCol, descCol, amountCol)) continue\n const date = formatDate(parts[dateCol])\n const desc = parts[descCol]\n const rawAmount = parseFloat(parts[amountCol])\n if (!date || isNaN(rawAmount)) continue\n // Normalize: negative = spend. Chase already negative, Amex needs negation\n const amount = isChaseStyle ? rawAmount : -rawAmount\n const id = makeId(sourceTag, date, desc, amount)\n try {\n await sqlite.execute({\n sql: `INSERT INTO transactions (id, source, date, description, amount, abs_amount, account_name, institution) VALUES (?, ?, ?, ?, ?, ?, ?, ?)`,\n args: [id, sourceTag, date, desc, amount, Math.abs(amount), accountName, institution],\n })\n inserted++\n rows.push({ date, desc, amount })\n affectedMonths.add(date.slice(0, 7))\n } catch {\n skipped++\n }\n }\n } else if (type === 'venmo') {\n let headerIdx = -1\n for (let i = 0; i < lines.length; i++) {\n if (lines[i].includes(',ID,Datetime,')) {\n headerIdx = i\n break\n }\n }\n if (headerIdx === -1)\n return Response.json({\n ok: false,\n error: 'Could not find Venmo CSV header',\n })\n for (let i = headerIdx + 1; i < lines.length; i++) {\n const parts = parseCSVLine(lines[i])\n if (parts.length < 9) continue\n const datetime = parts[2]\n const desc = parts[5] || parts[4]\n const amountStr = (parts[8] || '').replace(/[$ +,]/g, '')\n const amount = parseFloat(amountStr)\n if (!datetime || isNaN(amount)) continue\n const date = datetime.slice(0, 10)\n const id = makeId('venmo', date, desc, amount)\n const isIncome = amount > 0 ? 1 : 0\n try {\n await sqlite.execute({\n sql: \"INSERT INTO transactions (id, source, date, description, amount, abs_amount, account_name, institution, is_income) VALUES (?, 'csv_venmo', ?, ?, ?, ?, 'Venmo', 'Venmo', ?)\",\n args: [id, date, desc, amount, Math.abs(amount), isIncome],\n })\n inserted++\n rows.push({ date, desc, amount })\n affectedMonths.add(date.slice(0, 7))\n } catch {\n skipped++\n }\n }\n }\n for (const m of affectedMonths) await recomputeMonth(m)\n return Response.json({ ok: true, inserted, skipped, rows })\n } catch (e: any) {\n return Response.json({ ok: false, error: e.message })\n }\n}\n\nfunction makeId(source: string, date: string, desc: string, amount: number): string {\n const key = `${source}_${date}_${desc.trim().toLowerCase()}_${amount.toFixed(2)}`\n let h = 0\n for (let i = 0; i < key.length; i++) {\n h = ((h << 5) - h + key.charCodeAt(i)) | 0\n }\n return `csv_${source}_${date}_${Math.abs(h).toString(36)}`\n}\n\nfunction formatDate(d: string): string {\n if (d.includes('/')) {\n const [m, day, y] = d.split('/')\n return `${y}-${m.padStart(2, '0')}-${day.padStart(2, '0')}`\n }\n return d.slice(0, 10)\n}\n\nfunction parseCSVLine(line: string): string[] {\n const result: string[] = []\n let current = ''\n let inQuotes = false\n for (const ch of line) {\n if (ch === '\"') {\n inQuotes = !inQuotes\n } else if (ch === ',' && !inQuotes) {\n result.push(current.trim())\n current = ''\n } else {\n current += ch\n }\n }\n result.push(current.trim())\n return result\n}\n",[183],"df7c1d58-5ac2-11f1-8b07-ee650bb23af1","README.md","b4efaf8a-5aab-11f1-ae6c-ee650bb23af1",180,["D",1779992844459],"# Spend Agent\n\nAI-native personal finance tool. Connect your bank via [Teller](https://teller.io) or upload CSVs (Chase, Amex, Venmo), get AI-powered categorization, and see where your money goes.\n\n## How It Works\n\n![Architecture](https://kevinsuh--8ac5067e5aab11f1bb3cee650bb23af1.web.val.run/diagram.png?v=8)\n\nThe tool handles the boring part — syncing transactions from banks + Venmo and classifying them with AI. Once your data lands as structured rows, **you build** whatever you want on top.\n\n**The feedback loop is the key insight**: more data + prompt refinement = better classification over time.\n\n## What's Included\n\n- **Overview page** — Monthly burn at a glance. Stacked bar chart color-coded by category over 12 months. Click any month to drill down to individual transactions with AI reasoning.\n- **Ask Agent** — Chat with your financial data: _\"what did I spend on travel last quarter?\"_, _\"am I on track for my savings goal?\"_\n- **Settings** — Classification prompt editor, category management with color swatches, Teller bank connection, CSV upload.\n\n## Ideas After You Remix\n\nOnce your transactions are classified, the structured data is yours to build on:\n\n- **Find & cancel subscriptions** — \"Which recurring charges haven't I used in 3 months?\"\n- **Budget guardrails** — Cron that texts you when a category exceeds your monthly budget.\n- **Smart digest emails** — Weekly email: \"You spent $X this month, up 15% from last month.\"\n- **Custom automations** — Val Town crons + Claude + your categorized data = any workflow.\n\n## Setup (5 minutes)\n\n1. **Remix this project** on Val Town.\n2. **Set `ANTHROPIC_API_KEY`** in Settings > Environment Variables ([anthropic.com](https://anthropic.com)).\n3. **Open the dashboard.** Upload a CSV (Chase, Amex, Venmo) or connect a bank via Teller.\n4. Hit **Preview Classification** on the banner to categorize transactions with AI.\n\n### Optional: Teller bank sync\n\nSet `TELLER_APP_ID` to enable automatic bank sync. Get a free App ID from [teller.io](https://teller.io) (free up to 100 accounts, no credit card required). Sandbox mode works without mTLS certificates — just use the test credentials shown on the Settings page.\n\nFor production bank connections, you'll also need `TELLER_CERT` and `TELLER_KEY` (mTLS certificate PEMs from Teller's dashboard).\n\n## Classification Engine\n\n**AI classification** — Claude Haiku batch-classifies uncategorized transactions using Anthropic tool use with JSON schema enforcement. The `category` field is constrained to an `enum` — no hallucinated categories possible.\n\n**Interactive flow:**\n\n1. Click \"Review & Classify\" banner — drawer opens with unclassified transactions\n2. Review/edit the classification prompt — click \"Preview Classification\"\n3. Review results with color-coded category pills + confidence scores\n4. Edit prompt inline and \"Retry\" or \"Approve All\" to commit\n\n**Default categories:** Housing, Food, Transportation, Shopping, Entertainment, Health, Subscriptions, Bills, Travel, Investments, Transfers, Salary, Income, Gifts, Other\n\nCustomizable on the Settings page — add, rename, reorder, assign colors, or remove categories. Removing a category sends its transactions back to unclassified for re-review. Manual overrides on individual transactions are never re-classified.\n\n## Data Sources\n\n| Source | Provider | Cost | Best For |\n| -------------------- | -------- | ------------------------- | ----------------------- |\n| Banks & credit cards | Teller | Free (up to 100 accounts) | Chase, BofA, Amex, etc. |\n| CSV upload | Built-in | Free | Any bank, manual import |\n\n## Files\n\n| File | Type | Purpose |\n| -------------------- | ---------------- | ----------------------------------------------------- |\n| `app.ts` | http | Dashboard, settings, API routes |\n| `db.ts` | script | SQLite schema, config helpers, default categories |\n| `lib.ts` | script | Formatting, HTML shell, category colors, Settings JS |\n| `classify-engine.ts` | script | LLM classification engine (tool use with JSON schema) |\n| `classify-ui.ts` | script | Classification drawer UI |\n| `classify.ts` | interval (daily) | Thin cron wrapper — calls `runClassification()` |\n| `sync-engine.ts` | script | Teller mTLS sync logic |\n| `sync.ts` | interval (15min) | Thin cron wrapper — calls `runSync()` |\n\n## Cost\n\n- **Teller**: Free (up to 100 accounts)\n- **Claude Haiku**: ~$0.01/day for ~100 transactions\n- **Val Town**: Free tier works for personal use\n",[191],"pulls",0,"branches",1,"forks","envVars",{"_206":42,"_208":198},"connections",{"_206":198,"_207":198},"count","unfilledCount","emptyCount",{"_18":190,"_20":191,"_121":192,"_123":-5,"_124":71,"_125":193,"_127":38,"_22":243,"_68":42,"_130":142,"_132":195,"_134":26,"_135":-5,"_219":244,"_88":245},{"_18":182,"_20":183,"_121":36,"_123":-5,"_124":71,"_125":184,"_127":38,"_22":240,"_68":186,"_130":187,"_132":188,"_134":26,"_135":-5,"_219":241,"_88":242},{"_18":175,"_20":176,"_121":177,"_123":-5,"_124":71,"_125":69,"_127":38,"_22":237,"_68":179,"_130":142,"_132":180,"_134":26,"_135":-5,"_219":238,"_88":239},{"_18":167,"_20":168,"_121":169,"_123":-5,"_124":71,"_125":170,"_127":38,"_22":234,"_68":172,"_130":142,"_132":173,"_134":26,"_135":-5,"_219":235,"_88":236},{"_18":160,"_20":161,"_121":162,"_123":-5,"_124":71,"_125":163,"_127":38,"_22":231,"_68":157,"_130":131,"_132":165,"_134":26,"_135":-5,"_219":232,"_88":233},{"_18":152,"_20":153,"_121":154,"_123":-5,"_124":71,"_125":155,"_127":38,"_22":228,"_68":157,"_130":142,"_132":158,"_134":26,"_135":-5,"_219":229,"_88":230},{"_18":145,"_20":146,"_121":147,"_123":-5,"_124":71,"_125":148,"_127":38,"_22":225,"_68":42,"_130":142,"_132":150,"_134":26,"_135":-5,"_219":226,"_88":227},{"_18":137,"_20":138,"_121":139,"_123":-5,"_124":71,"_125":140,"_127":38,"_22":222,"_68":42,"_130":142,"_132":143,"_134":26,"_135":-5,"_219":223,"_88":224},{"_18":119,"_20":120,"_121":122,"_123":-5,"_124":71,"_125":126,"_127":38,"_22":218,"_68":129,"_130":131,"_132":133,"_134":26,"_135":-5,"_219":220,"_88":221},["D",1779741957609],"children",[],[120],["D",1779993915129],[],[138],["D",1779995223579],[],[146],["D",1779998124644],[],[153],["D",1779903197447],[],[161],["D",1779901810486],[],[168],["D",1780009107326],[],[176],["D",1779998557094],[],[183],["D",1779992844459],[],[191],["D",1779740379107],["D",1789504908499],{"_16":17,"_86":82},{"_16":17,"_80":-5,"_81":82,"_33":34,"_83":84,"_85":26},{"_68":69,"_20":70,"_18":71,"_72":19,"_73":-5,"_74":-5,"_75":-5,"_76":-5,"_77":-5,"_78":21,"_79":17},"ctx",{"_61":-5,"_107":284,"_285":286,"_287":288,"_289":26,"_290":26,"_291":26,"_292":38,"_293":-5,"_294":-7,"_295":-7,"_296":198,"_297":26,"_298":26,"_299":-5,"_300":26,"_301":26,"_302":26,"_303":-5},"ENV",{"_255":256,"_257":258,"_259":70,"_260":261,"_262":263,"_264":265,"_266":267,"_268":269,"_270":271,"_272":273,"_274":275,"_276":277,"_278":279,"_280":281,"_282":283},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

Username	`username` (or `otp` / `challenge` for MFA flows)
Password	`password`
OTP code	`0000` (if username = `otp`)
Challenge	`blue` (if username = `challenge`)

Search