Nostrix compiles a cross-sectional scoring engine, a regime classifier, out-of-sample verification and REST/MCP endpoints into a single local binary. No notebook rebuilds, no cloud key to start, and your data never leaves your machine.
free for students & researchers · Pro Trader $99/mo · Pro Office $2,500/mo — see all three
Most research tooling is a pile of notebooks, stale CSVs and vendor calls that break on Monday morning. Nostrix is the opposite: one compiled, local, auditable engine you can put in a pipeline and defend line by line.
Nothing is uploaded. Run it in CI, on an air-gapped box, or against a dataset your compliance team will never let you send to a vendor.
Same file in, same score out — seeded Monte-Carlo, pinned constants, explicit weights. Every number traces back to μ/σ/Δ/Γ you can audit.
One command ranks a whole universe through a compiled engine — instead of re-writing the notebook, re-fetching, and re-tuning for the third time today.
`nf verify` replays every saved ranking snapshot against realised prices: hit rate, average, best and worst per snapshot. Claims stop being claims.
Every one of these is a single line in a terminal — scriptable, CI-able, and equally at home in a research run or a pre-open checklist.
The flagship. Layer 1 boundaries (Q/V/G/H) → Layer 2 SDE (μ σ Δ Γ) → Layer 3 gates (P, K) plus risk index, upside potential and alpha — with sector concentration caps.
Ten layers on one ticker: technicals, financials, analyst, consensus, ensemble prediction, peer set and synthesis — one call, no dashboard tab-sprawl.
11-dimension regime classifier with dynamic weights, and `nf decide` for a rank × score × predict consensus with quarter-Kelly sizing, stop and max-risk.
Loads saved rank history, re-prices the top 5 predictions and reports win rate, average return, best and worst per snapshot — real, not back-filled.
A local REST server exposing 22 functions as JSON/CSV. `GET /api/call.csv?fn=rank` drops straight into Sheets, pandas or Excel. No AI endpoint exposed.
16 MCP tools for Claude Desktop / Claude Code, anchored to verbatim CLI output. Bad tickers return `isError`; destructive commands are refused outright.
The whole reason quants trust it: nothing is a mystery box. Layer 3 can always be traced back to Layer 2, and Layer 2 back to the columns you tagged yourself.
Tag any column inline — `Revenue:V+`, `Margin:Q1+:P3`. Dimensions Q/V/G/H, numbered sub-metrics auto-balance, P-tags set two levels of priority.
μ = deviation from the midpoint, σ = dispersion, Δ = the quality–growth gap, Γ = compound interaction. Cross-sectional, computed from the entity's own boundary.
Gates aggregate into P (probability) and K (Kelly fraction) — layered with risk index, upside potential and agreement into a 0–100 score and a direction.
The same engine in all three — what changes is how much of it you get and who it is built for. Start free as a student, move up when your P&L does, or run it across the whole office.
Free forever — the full analysis CLI on your own machine. No key, no signup, no time limit.
The research desk for retail quants: C++ ranking, consensus verdicts, scanning, dashboards, API and MCP.
Proprietary data, custom engines, global feeds, seats and a support contract.
Student · free build, no key → Trader · Pro build → Office · Ultra edition + nd pipeline + custom engine
curl -fsSL https://nostrix-finance.vercel.app/nf-pro | bashDetects your OS, downloads the right binary, puts nf on your PATH.
nf scripter yfBundled, ~110 ms. Or point NOSTRIX_SCRIPTS at your own script — SEC EDGAR, FRED, a vendor API, your scraper. BYOS by design.
nf q AAPLThen nf rank --top 50, nf decide AAPL, or nf mcp install to wire it into Claude.
Pro — everything in Free plus the C++ multi-factor ranker, decide/seek, dashboards, the REST API and MCP for Claude. Ships as the default download.
curl -fsSL https://nostrix-finance.vercel.app/nf-pro | bash