Your bank shows a balance. Rent, a credit-card statement, three subscriptions and four days until payday show something very different. Tide answers the one question that actually matters in the moment — what's safe to spend today? It pulls every checking account, savings account and credit card into a single daily view, learns where the money really goes, and keeps a payday-aware runway you can trust — with an AI coach, grounded in your real accounts, for the questions a dashboard can't answer.
A solo build — sparked by my own too-many-apps, what's-left-till-payday problem · React Native · Expo · Claude · AWS Bedrock · AI evals · ~$46/mo AWS budget
Three frustrations kept colliding for me every single day — and no single app solved all three at once.
A few checking and savings accounts. A couple of credit cards. Each living in its own app, behind its own login. Just seeing where I actually stood meant opening five apps every morning — and I'm someone who likes to check daily.
The number in the account already belongs to rent, pending bills and that card balance. "What can I spend?" is a different — and usually smaller — number than "what's my balance?", and it moves with the paycheck cycle, not the calendar week.
Every app categorizes differently, so spending never added up across accounts. With money flowing through several places at once, "where did it actually go?" simply had no trustworthy answer.
The honest origin: this began as my own problem — too many accounts to track, a balance that lied about what was really mine to spend, and spending I couldn't see clearly. The more I dug in, the clearer it got that plenty of people on a steady paycheck live the exact same gap. So I built it to solve that well for one person first, then opened up the use cases that travel — to everyone in the same situation.
Sharp focus beats broad coverage. Tide goes deep for one segment instead of shallow for everyone — and is just as deliberate about who it's not for yet.
Salaried professionals on a regular paycheck who juggle a few accounts and cards, carry the odd balance and subscription, and still feel a low-grade uncertainty every time they tap a card. Not broke — just never quite sure what's truly safe to spend before the next paycheck lands.
Not these people
Not these jobs
Some of these are on the roadmap — thin-history and multi-job robustness especially. But naming what Tide deliberately is not doing is exactly what lets it be genuinely good at the one thing it is.
The customer's words, not the feature list — each tied to the outcome it has to deliver.
"When my money's spread across five accounts and three cards, I want one place that shows me where I stand — every morning, in a single glance."
Outcome Every account and card aggregated into one daily view. No more app-hopping just to see the truth.
"When I'm about to tap my card, I want to know I won't wreck rent or a bill before payday — so I can spend without the quiet dread."
Outcome A payday-aware Safe-to-Spend number that already nets out upcoming bills and credit-card debt.
"When I look back at the month, I want to actually understand where my money went — consistently, across every account."
Outcome One categorization engine that learns, instead of five apps disagreeing with each other.
"When I have a money question a chart can't answer, I want to just ask — and get a straight answer grounded in my real accounts."
Outcome An AI coach that reasons over your actual data with a defined tool-set — not generic tips.
Each job above only holds up if something underneath earns it. I designed from the outcome the customer wants and built back toward the plumbing — never the other way around.
The category is full of genuinely good apps. Tide deliberately doesn't try to be any of them — it occupies the gap between them.
Comprehensive budgets, net-worth tracking and investment aggregation built for couples and whole households.
Tide's wedge: not a net-worth tracker — a daily, in-the-moment "what's safe to spend before payday" for one earner.
Gorgeous design and strong categorization of where your money has already gone.
Tide's wedge: forward-looking runway, not a backward-looking report — it answers what's left, not just what was spent.
A chatty, gamified AI persona for a younger audience, with nudges and cash advances.
Tide's wedge: a coach grounded in your actual accounts and gated by evals — accuracy and trust over personality.
Hunts down forgotten subscriptions, negotiates your bills down, and flags spending you're quietly leaking.
Tide's wedge: not a leak-plugger or bill negotiator — the daily "what's safe to spend before payday" number, whatever your subscriptions are.
For salaried earners juggling several accounts and cards, Tide is the one daily view that turns scattered balances into a single honest number — what's safe to spend before your next paycheck — and lets you ask why, with AI you can actually trust.
The story above is the why. These nine interactive views are the how — the system topology, the data flows, the AI pipeline, the evals, and the runway math behind every claim. Click any card to explore.
Interactive system topology showing all 13 components. Expo Dev Build with native modules, Native Plaid Link SDK, XHR-based SSE streaming. Click nodes for details.
13 NODES · 17 EDGES ⇄Step-by-step animated visualizations of 11 critical data flows: auth, Native Plaid Link bank connections, AI chat via XHR SSE, transaction sync, the payday-aware Safe to Spend daily-pace ledger (with credit card debt), spending insights, bill predictions, and more. Evening Reflection removed (MVP+); Spending Challenges removed 2026-06-10; Smart Pulse removed 2026-06-12.
11 FLOWS · ANIMATED ◈Multi-agent architecture: Haiku intent classifier, Sonnet Tide Coach (8 tools), Haiku merchant classifier (fills Plaid's Software & Tech / Subscriptions gap, cached per-merchant on merchants.ai_category_id). Home-parity Safe to Spend tool. Full chat persistence with sync DB fallback.
Rebuilt Tide Coach’s evals the Hamel Husain & Shreya Shankar way: 100 scenarios from a job × stage grid, asked of the live Coach on real bank data and traced end to end in Arize. I graded the first answers myself and aligned an AI judge to my grades on a held-out set, then grouped every failure by its first upstream cause — 45 of 53 were prompt, tool, engine or data bugs, not model reasoning. Each fix was proven before it shipped: prompt changes on frozen replays (same recorded data, only the change differs), tool and engine changes on live re-runs, every grade blind. Code checks now catch invented figures for free; the scripted suite keeps running until the new judges are certified.
100 REAL-DATA SCENARIOS · 14 FAILURE TYPES · FROZEN REPLAYS · BLIND JUDGE ☰Every nightly run, every test, graded — failures included. The misses are what decide the next tests and prompt fixes.
NIGHTLY · SINCE 18 SEP 2026 · EVERY TEST ⛊Documented iteration history for every major feature. Each version entry shows what changed, what motivated it, what we learned, and what we kept. Start with Safe to Spend — 23 versions across five weeks.
1 FEATURE · 23 VERSIONS · MORE SOON ΣClick-to-expand component breakdown of the whole runway formula. Each component card shows source, time-window semantics, code reference, and a worked example. Pairs with the Typical Spending math spec for the forecast walkthrough.
7 COMPONENTS · 4 TABS · OPEN QUESTIONS LISTED 📋Every idea, scored and bucketed. 122 features across 23 epics, split into the June 23 MVP demo slice and a prioritized P1–P4 roadmap, with shipped/partial status badges so progress stays visible. Drag-to-reorder to explore prioritization trade-offs.
23 EPICS · 122 FEATURES · MVP + ROADMAP ⚙How the Tide repo runs as a Claude Code workspace: session lifecycle, monorepo map, the hooks vs conventions vs skills vs agents vs routines layers, MCP servers, and the self-maintenance loops (role detection, doc-sync, portfolio-mirror sync, LEARNINGS, nightly doc-drift, CI/evals).
5 SECTIONS · HARNESS MAP