One install, one way in: the web terminal, served by your own machine and driving a local engine, license and data — your browser is the front-end, and a native desktop window wrapping the same terminal is built for a future installer. It runs the whole research-to-execution stack locally (execution is paper-only in v1): an on-device AI analyst, deflated backtesting, an options lab, flow and microstructure signals, and a hard risk kernel. Your keys, your data, and your models never leave the machine — and every performance number ships with its deflated statistics.
Each of these is a real, shipping module — not a roadmap promise.
A local large language model runs entirely on your GPU or CPU — no cloud round-trip, no prompt logging, no data ever leaving the machine. It reasons over your signals and positions to draft plain-language takes, and stands for a weighted seat inside the multi-bot ensemble rather than acting as an oracle — a seat whose weight is earned against resolved outcomes and measured 0.00 on 2026-09-05. How the analyst works →
Test strategies written in Pine or Python against your own history. Results are always reported as raw SR | PSR | DSR with the trial count N in view, so multiple-testing inflation is deflated away instead of hidden. If a Sharpe survives the deflation, you'll know — and if it doesn't, you'll know that too.
Real Black-Scholes Greeks, risk-neutral probability-of-profit, and a GEX / dealer-positioning view built from the chain. A defined-risk recommender proposes spreads and condors sized to a fixed max loss, with a chain-age stamp so you always know how stale the quotes are.
FINRA ATS weekly volumes plus TRF daily prints feed a calibrated microstructure rating — z-scores and deltas, not made-up numbers. When a data source needs a key you don't have, the module says so instead of fabricating a signal.
Trace a company's suppliers and customers across tier-N relationships on an interactive graph, then quantify concentration with an HHI score and single-source flags. Data is file-driven and sourced from public filings — no invented edges.
Separate short-, medium-, and long-horizon models produce predictions wrapped in conformal prediction bands, so you get a calibrated interval rather than a false-precision point estimate. A self-grading feedback loop scores each call against the real outcome and re-weights the ensemble over time.
A unified kill switch and position, per-trade loss, and drawdown limits sit on every order path. One click halts new orders; the limits gate execution before anything reaches the paper book or your broker's paper endpoint — all trading is paper in this version. Nothing trades until you explicitly arm it.
Surface the high-impact figures and scheduled events that move your watchlist — earnings, macro prints, and known catalysts — so you're not blindsided by a date on the calendar. It flags what's coming; it does not tell you how to trade it.
A built-in SQLite manager with a read-only query console lets you inspect every table the platform keeps — signals, the feedback ledger, cached market data — directly. Your research history is yours to browse, export, and audit.
This is the app: a single-page terminal served by your own machine at 127.0.0.1:8000, local-only and behind the same license. One engine, one set of databases, one trial ledger — the numbers on screen are the only numbers there are, with the same deflated statistics everywhere. A native desktop window that wraps this same terminal is built and will ship in a later installer.
A native Windows install (macOS and Linux builds coming soon) that bundles its own dependencies and runs offline-capable. It connects with your broker and data keys (Alpaca, Finnhub, Tradier and more), stored locally — plus NinjaTrader over its local ATI file bridge and FRED, which needs no credential — and licenses via an offline HMAC-signed key. It phones home to nothing.
A screen count is just a number; here is how the screens group into the desks you'd actually sit at. Real captures of most of these are on the tour.
The Dashboard stamps every panel with the price basis it used — movers, breadth, sectors, the economic calendar, your watchlist. The Morning Brief assembles what changed overnight from your own configured feeds, and Alerts watch the levels you set. Where a feed hasn't earned enough data, the panel says so instead of filling in.
The Options Lab computes real Black-Scholes Greeks and proposes defined-risk structures sized to a fixed max loss, with a chain-age stamp on the quotes it used. GEX Analysis reads dealer positioning out of the chain; the Flow Scanner and 0DTE screens watch how premium is leaning; the Options Bot paper lane lets a strategy trade it — simulated, in its own ledger.
Write a strategy in the Pine subset or restricted Python, walk it forward, and let Experiments compare variants with autocorrelation-robust paired tests instead of eyeballed equity curves. The Validation Center walks the gates in order on your own install's data, and the Edge Report ends the argument: a signed, deflated verdict with the breakeven N where the edge dies.
The Portfolio Optimizer allocates by Hierarchical Risk Parity with Kelly-based sizing; the Risk Engine prices VaR 95, expected shortfall, and Cornish-Fisher VaR 99 on what you actually hold; the Stress Tester replays bad regimes against your book. A safe-mode ladder and a one-click kill switch sit above every order path — and a risk figure that isn't measured yet is withheld with its reason, never shown as 0.00.
13F Holdings and SEC Filings read straight from EDGAR; the Insider Desk tracks Form 3/4/5 filings with a deep link to the primary EDGAR document on every row; Dark Pool builds its ratings from FINRA ATS weekly files and TRF daily prints, sourced and dated on screen. Supply Chain maps tier-N supplier and customer dependencies with concentration scores, and Event Studies let you test what a catalyst has historically been worth.
Neural Analysis shows the champion model and the promotion record it survived; the Refusal Ledger lists every candidate that didn't, with reasons. The AI Analyst writes grounded desk notes — cite a real number or abstain — while the LLM Monitor grades its calls against outcomes and the Model Lab lets you register, activate, and roll back the local models it all runs on.
Not advice, and honest about it. QuantSuite is research and paper-trading software — not a broker-dealer, not a registered investment adviser, and not a source of personalized investment advice. Every performance figure is deflated and labeled; nothing trades, even on paper, until you arm it. Trading involves substantial risk of loss.
One license. Every module above. Your keys, your data, your models — local by default.