open source · Claude Code skill · Python library

Quant Garage.Pro research.No terminal.

Analyst-grade market research used to live behind expensive professional terminals. Now it runs in your Claude, or a Python notebook, grounded in live data and silent when the numbers won’t support a take.

Thirteen one-command workflows chain multiple tools into a single market briefing. One prompt returns the regime, the risk, the upcoming earnings, the macro calendar, and the sector rotation. The read a research analyst would put together at 8am, computed on demand.

Trade like a pro. Without the terminal.

42tools·13workflows·1key·MIT
market-regime09:31 EDT
risk_oncomposite
SPY trend
uptrend_strong
VIX pct rank
14th
breadth > 50d
9 / 11
reasons: 20/50/200 stacked, VIX 12.4, XLK & XLC leading.
earnings-drilldown NVDAQ1 preview
implied move
±7.2%
realized 8q
±5.4%
take: straddle rich vs realized. PEAD t-stat 2.1, sign conditional on upside.
risk-reportvar share
ALLO
66%
NVDA
12%
MSFT
8%
GOOGL
7%
AMZN
4%
flag: ALLO concentration > 50%. rebalancer suggests trim $66k.
historical-analog-finderk=20, SPY
fwd 252d return dist
-20%median +14.3%+40%
hit rate 252d: 75% · analogs: mid-2024, late-2021.
Four tool outputs. Same key. Same audit trail.
01What it is

The full research desk, one prompt away.

Forty-two tools that each answer one analyst question well. Earnings previews. Risk reports. Factor research. Options flow. Corporate actions. Macro calendars. Thirteen workflows chain them into complete briefings, so one prompt returns the whole read.

Every figure carries the endpoint and timestamp it came from. Every take shows its work. Ask from Claude, import into a Python notebook, or wire it into a cron job. Same computation, same audit trail, same answer.

Built for
AI · agents

LLM and agent developers.

Every tool returns JSON matching a schema. Drop it into your tool-use loop and the model gets structured research instead of hallucinated finance. Ships as Claude Code skills; works with any tool-use LLM.

Claude Code

Traders who work in Claude.

Ask what you'd ask a colleague. "Preview NVDA earnings." "Should I trim ALLO?" The skill loads whenever the conversation touches markets. Sell-side output up top, cite trail underneath every take.

Retail

Investors with a real book.

$50K to a few million. No Bloomberg, no time to babysit a spreadsheet. Get sell-side-quality briefings on the names you actually hold: morning brief, earnings preview, portfolio review.

Fintech · dev

Developers building on Massive.

Reference implementation for the Massive API. Retry logic, fallback chains, timezone handling, audit-trail format. Read the source or fork it and ship your own.

02In Claude Code

Prompt an analyst who reads your book.

The skills load automatically whenever the conversation touches markets. No slash commands, no invocation grammar. Ask what you’d ask an analyst on your desk. Claude routes to the right tool, pulls the numbers, and writes the note.

preview NVDA earningsearnings-drilldown
should I trim ALLO?risk-report
what's the tape doing?market-regime
review my bookportfolio-review
claude codequant-garage
you
preview NVDA earnings
claude
running earnings-drilldown NVDA...
implied ±7.2% vs realized 8q avg ±5.4%. PEAD t-stat 2.1, positive-sign conditional.
take: straddle rich vs realized. sell-side ranks pass on the print.
4 API calls · 340ms · fx.polygon.io/v3/reference/tickers/NVDA
03In Python
notebookpython 3.11
# Cell 1 — the same tool Claude calls, imported.
from quant_garage.portfolio_review import run
 
report = run(book_csv="book.csv")
print(report["headline"])
 
# regime: risk_on (VIX 14th pct, breadth 9/11)
# variance top: ALLO 66% ⚠ flag
# next earnings: NVDA Thu 4:20p ET
# next macro: FOMC Wed 2:00p ET

Or skip Claude. Call the same function in Python.

Every skill is an importable Python function that returns JSON. Same client, same audit trail, same fallback chain as the Claude path. Drop it wherever you already write code.

Jupyter notebooks — return JSON, plot with pandas
Slack bots — post the morning brief at 8am
Cron jobs — daily portfolio-review after close
Dashboards — Streamlit, Retool, anything that renders JSON
04Thirteen one-command workflows

Start with a workflow. Get the whole read in one call.

Each one picks the right tools, chains them in order, and hands you back a single briefing. Point it at your watchlist or a single ticker.

skills/*
deep dive~90s
portfolio-review
The full book review in one call. Chains market-regime, sector rotation, risk report, earnings blackout, macro calendar, 8-K scanner, rebalancer.
weekly~30s
weekly-brief
Sunday-night watchlist prep. Regime, rotation, macro calendar, earnings blackout across a 7-day window. Frames the week.
daily~15s
morning-brief
60-second daily open. Regime, macro today plus tomorrow, news scan per watchlist ticker. Cron it at 8am.
pre-trade~30s
preflight-trade
Before you hit execute. Ticker plus intent. Technical read, earnings blackout, news, corporate actions. Verdict: go, wait, review.
earnings~2 min
earnings-week-prep
For weeks where four of your names print in five days. Earnings blackout on the watchlist, then drilldown plus technical read for the top prints.
analog~20s
historical-comparison
Twin decision support. Event study on the specific event plus historical analogs on the market regime. Two anchors, not one.
discovery~30s
scan-and-frame
Regime-framed idea generation. Regime context, universe screen, relative-strength rank on the top N. Factor pass optional.
single-name~10s
stock-one-pager
Retail-tier single-name card. Technical read plus earnings blackout plus regime, translated into plain language. Every claim gated to data.
05Forty-two primitives

The building blocks. Compose your own workflow.

Each is a standalone run() → dict with the same client, timezone handling, audit-trail format, and significance thresholds.

Earnings3
earnings-drilldownevent-studyearnings-blackout
Equity research & valuation3
pitch-compsvaluation-sanity-checkposition-sizer
Quant & screening9
technical-briefinguniverse-builderfactor-researchrelative-strengthpairs-scannerhurst-exponentchange-point-detectorsignal-decayrough-vol-forecast
Market context4
market-regimesector-rotation-signalmacro-event-calendarhistorical-analog-finder
Trading & execution5
options-flowzero-dte-gammanews-scannerslippage-costoptions-structure-analyzer
Risk & ops7
portfolio-markrisk-reportmc-portfolio-simulatorportfolio-rebalancercorporate-actions-scannercorp-actions-reconcilert+1-settlement-prep
Filings & ownership7
8-k-scannerrisk-factor-deltafiling-sentimentinsider-flowmanager-portfolio-diffguidance-trackeranalyst-tracker
Backtesting & crypto2
backtest-data-prepcrypto-vol-scanner
06Under the hood

The quant methods most investors haven’t seen.

A handful of these tools implement published academic research from 2007 to 2025 that individual investors have never had access to. Four worth calling out.

quantAdams & MacKay, 2007
change-point-detector
Bayesian Online Change-Point Detection on daily log returns. Flags the exact dates where the return-generating distribution shifted, with posterior confidence and per-segment vol.
Live resultLive on ALLO: detected a regime break on 2026-04-14 with 99.65% confidence. That was the exact day of the ALPHA3 trial futility announcement.
quantBayer-Friz-Gatheral, 2016
rough-vol-forecast
Under rough volatility, realized vol scales as h^H with H around 0.14, much slower than sqrt(t). Ships the rough forecast alongside Brownian and EWMA for direct comparison.
Live resultOn SPY: 120-day vol drops from 11.66% (traditional) to 2.08% (rough), a 5x reduction that reshapes how you size long-horizon scenarios.
optionsBaltussen-Terhorst-Van Vliet, 2024
zero-dte-gamma
Estimates net dealer gamma exposure for same-day-expiry SPY/SPX/QQQ/IWM options. Identifies pin strikes and gamma-regime flip levels driving intraday moves.
Live resultLive SPY 2026-07-13 expiry: dealers net short $1.25B gamma at $759 pins with 9,412 open interest. Expect intraday moves to accelerate rather than mean-revert.
quantIsrael-Moskowitz-Ross, 2024
signal-decay
Rolling 63-day IC vs 5-day forward returns over a 5-year window. Fits exponential decay to the IC series and reports the half-life plus a full performance tearsheet.
Live resultSPY mean-reversion IC flipped from +0.06 early quarter to -0.16 recent quarter. The classic short-horizon effect has completely reversed sign this cycle.
07Why the layer exists

Every take is computed. Every number cites its call.

The whole difference between a research tool and a chatbot with a market-data plugin is that these tools refuse to fabricate when the data isn’t there. They surface the endpoint and timestamp for each figure so you can retrace the reasoning.

LLMs and agents work better on top of this surface than under it. Humans read a briefing, agents read a schema, both anchored to the same live citation trail.

What comes back with every take
take
Straddle rich vs realized. PEAD t-stat 2.1.
supporting
Implied ±7.2% vs realized 8q avg ±5.4%. Sign conditional on positive reaction.
endpoint
v3/reference/tickers/NVDA
timestamp
2026-07-13T09:31:07-04:00
sample
32 quarterly prints (8 years)
Any claim on the page can be verified against a live call in seconds. Same fields ship with every tool response.
08What you need to run it

You need a Massive API key. Free works for most of it.

Every install needs a key. The Basic tier is free and runs 21 tools plus most workflows end to end. Stocks Starter at $29/month opens 39 of 42 tools and every workflow. Add-ons matter only for options data, live streaming, and Benzinga fundamentals.

Basic (free)
$0
21 tools + most workflows end to end
5 calls/min, end-of-day data, SEC EDGAR fallback for earnings
Stocks Starter
$29/mo
39 of 42 tools, every workflow composite
unlimited rate, 15-min delayed real-time, bulk grouped-aggregates
Options Developer
$79/mo add-on
options-flow, options-structure-analyzer, full earnings-drilldown
Stocks Advanced
$199/mo
Live-mode portfolio-mark on the WebSocket feed
Get a key at massive.com/pricing. The PLAN-MATRIX file in the repo maps every tool to the exact plan and add-ons it needs.
09Install

Two ways to install. One key.

Grab a Massive key first. Then install the Python package, the Claude Code plugin, or both. Same functions, same audit trail, same JSON, whichever way you call them.

  • Works in a Jupyter notebook, a Slack bot, or a cron job
  • Same code from Python, Claude Code, or any tool-use LLM
  • SEC EDGAR fallback for earnings when Benzinga isn't wired
terminalzsh
# 1. Get a Massive API key (free tier works)
# massive.com/pricing
$export MASSIVE_API_KEY=your_key_here
 
# 2a. Use it from Python
$pip install quant-garage
$python -m quant_garage.portfolio_review book.csv
 
# 2b. Or install the Claude Code plugin
$claude plugin marketplace add https://github.com/rgourley/quant-garage
$/plugin install quant-garage
 
# Then ask Claude in plain language
>review my book and flag anything I should look at
>preview NVDA earnings
>what's the tape doing?
MIT · by Robert Gourley

Built in the garage, not the trading floor.