For prop trading desks
Where whales do their research
Play in the next league.
AI-native research workbench for traders.
Play in the next league.
AI-native research workbench for traders.
FinWhale is a smart brain layer that runs your desk's research and keeps everything it learns — the memory, the tools, and the orchestration underneath. Run it two ways:
A research agent on the web, with the brain-memory layer behind it. Nothing to install; your desk logs in and works.
Already living in Claude Code, Codex, or OpenClaw? FinWhale rides on top as a rich, robust memory and toolset — point-in-time data, durable overnight jobs, and the brain — inside the harness your quants already use.
Either way, one brain — and it's proactive. It doesn't wait to be asked. It comes to you.
Every call, thesis, and setup becomes durable memory — typed by how long it stays true, keyed to the names and sectors it touches.
Every forward call is resolved against the market. You see its real hit rate by setup and by regime — graded, not guessed.
Calls carry a shelf life. A setup that worked last season but inverted this one gets flagged, not quietly re-served as live.
It recommends the next move and stands up live watchers on your approval — and retires them when the edge decays.
The brain learns from every session and improves itself, so the desk it hands you tomorrow is sharper than the one you used today.
A general chatbot will search the web and sound confident — but the answer comes off whatever page it landed on, verified or not. FinWhale answers from real market data and shows you exactly where each number came from.
“What’s the earnings calendar this week — and who’s already reported?”
A web search stitched from whatever pages turn up — some stale, some plain wrong, and no way to tell which.
This week’s actual confirmed reports, with the source for every date.
“Who actually trades with this name — not just its sector label?”
A canned peer group from a static sector bucket — the same list everyone gets, blind to how these names move together now.
Peers ranked by how they’ve genuinely co-moved and led each other — with the lag and the strength.
“When does this IPO’s lock-up unwind — and how much unlocks?”
A number scraped from some article — no share count, and no primary filing to check it against.
The real expiry date and the share volume unlocking — with the filing it came from.
“Who gets hit most if the Strait of Hormuz closes?”
A plausible list of “oil names” — no revenue weights, no split between who’s hurt on demand versus on cost.
Names ranked by real revenue exposure to the region — demand and cost — traceable to filings.
“Who are AAPL’s top suppliers — actually, today?”
A web search that surfaces a years-old list — and no way for it to tell the list went out of date.
Current suppliers and how material each one is (“X is Z% of Y”) — actual, not years out of date.
Real answers from data you can check — not a confident guess off an unverified page.
When a quick answer isn’t enough, hand the brain a hypothesis and it runs a real backtest — overnight if it has to. The hard part was never the compute; it’s not lying to yourself. FinWhale is built for both.
Every test runs on point-in-time data — the brain only sees what was knowable on each day, so a result can’t be inflated by information that didn’t exist yet.
Tests that take hours run as durable jobs. Kick one off, close your laptop, and the result is waiting — it survives the overnight run.
The agent walks you through the setup — universe, dates, fills, what counts as a win — so the question is well-posed before a single number is computed.
Fills, gaps, and halts are modeled, not wished away. Every figure traces back to the data it came from — trust the number, or throw it out for a reason.
FinWhale captures that edge as it happens and compounds it, session after session — so the desk gets measurably better instead of starting from zero every morning. You still call the trade.
The setups, the names, the way each trader sizes — that's personal IP. FinWhale is built to protect it, not pool it.
Each trader's research, calls, and brain are theirs alone. The brain learns one trader and never carries what it learned to another.
Two traders pool a setup only if both choose to — and only with each other. Never the whole floor. Never us. No cross-desk model.
Your data stays in your environment. Leave, and you walk out with everything you built — the brain, the watchers, the signals.
We don't own the agent loop. We run on the best models there are and get stronger as they get smarter.
What a smarter model still can't manufacture is what we build:
The gains from AI scale with the model (DeepMind, “From AGI to ASI,” 2026) — and a non-stationary, partially-observed market is exactly where scale alone stops being enough.
Forward-deployed and founder-led. We audit where your desk repeats itself, then build the automations into the brain — past what ships in the standard product, fitted to how your traders actually work. Two to three desks per quarter.
I'll show you the brain on real market data and how the partnership would work for your desk.