Most financial forecasting on the internet rounds to two things: confident calls without scoring, and scored calls without conviction. The first is entertainment dressed as analysis. The second is so over-hedged that it's useless. We believed there was a third option — confident, scored, and improvable — and that AI agents made it cheap enough to actually try.
So we built FutureDaily. Every morning at 8:00 IST a swarm of seven independent agents produces a USD/INR forecast for the next day, the next month, and the next three years. Every evening we mark yesterday's call to market and write down what we got wrong. The scorecard is the front page, not the footer. The 365-day cone is wide on purpose.
USD/INR is the first asset because it's the one we know best and care about most. More are coming, but only after each new asset earns its way in through 90 days of paper-trading against an internal Brier-score threshold. We earn the right to forecast each one.
Every line of code, every prompt, every architectural decision is documented in the open repo. The daily build log lives on X. The methodology page is the spec. The article you read every morning is the same JSON our internal Slack reads. There is nothing behind a paywall because there is nothing to hide.
Forex traders, retail investors, importers and exporters, treasurers managing currency exposure, family offices, NRIs sending remittances, students learning markets, and anyone who has ever read a research note and wondered "but were they right last time?". We owe each of you the same thing: a forecast that is checkable.
We are not investment advice. We are not a managed product. We are not selling signals, courses, or proprietary indicators. FutureDaily is research output published every morning. What you do with it is up to you, and how it lands the next day is on us.
Forecasting runs on Kimi K2's agentic mode for the swarm reasoning, orchestrated through a lightweight Next.js 15 + Supabase backend. Engineering is done with Claude. Charts are Plotly. Hosting is Vercel. Cron is GitHub Actions. Email is Buttondown. Errors are Sentry. Analytics are Plausible. Everything we use is paid for, not sponsored.
We read every reply to the daily email. We read every DM on X. If you spot a methodology bug, an arithmetic error, or a forecasted event we missed, we want to know — and we'll publish the correction the next morning, named and dated.