VQuant
Institutional-grade financial intelligence, AI-powered
A Claude-driven financial analysis platform pairing 265+ data endpoints with a quantitative Python engine and streaming research reports.
265+
financial data endpoints
213
Claude agent tools
17
Python scientific libraries
6
selectable Claude models
36,000+
lines of code
41
tests passing
Overview
VibeQuant is a production-grade, full-stack financial analysis platform. A Claude AI agent — six selectable models with up to 1M tokens of context, Fable 5 by default — wields 213 specialized financial tools over 265+ data endpoints, a quantitative Python engine, and multi-source web research, delivering institutional-grade insight through a real-time streaming conversational interface.
The data layer covers company fundamentals, market data, technical indicators, analyst ratings, insider and institutional activity, SEC filings, macro indicators, ETFs, forex, and crypto. The quant engine runs Monte Carlo simulation, Black-Scholes with full Greeks, GARCH volatility, VaR, mean-variance portfolio optimization, and arbitrary Python across 17 scientific libraries.
Finished analyses export in one click to PDF, Word, and LaTeX Beamer slides, and a searchable community showcase publishes hundreds of analyses with the exact tool-chain the agent used on every card. Built in TypeScript with React 18, Express, Drizzle ORM, and Zod validation — live at www.vquant.ai.
Key Features
Claude agent with 1M context
Six selectable Claude models with adaptive extended thinking shown live in a collapsible reasoning widget, plus vision support for charts and documents.
265+ financial data endpoints
Fundamentals, market data, technical indicators, analyst ratings, insider and Congressional trading, SEC filings, macro, ETFs, forex, and crypto via FMP.
Quantitative Python engine
Monte Carlo with 10k+ paths, Black-Scholes with full Greeks, GARCH(1,1) forecasting, three flavors of VaR, and mean-variance portfolio optimization.
Arbitrary Python execution
The agent writes and runs custom analysis code with access to 17 scientific libraries — numpy, pandas, scipy, arch, cvxpy, quantstats, and more — plus live FMP data.
Multi-source web research
Tavily AI search, Firecrawl scraping with PDF extraction up to 750 pages, Exa semantic search, and SerpAPI Google surfaces feed the agent's research.
Real-time streaming interface
Server-Sent Events stream responses instantly, with intelligent tool batching of 5 tools per batch and per-session token and cost tracking.
One-click report export
Puppeteer-rendered PDFs, Pandoc Word documents, LaTeX Beamer slide decks, CSV/XLSX/JSON data files, and shareable report links with unique IDs.
Community showcase
Hundreds of public analyses, searchable, each card showing the exact tool-chain the agent used to produce the report.
How It Works
Streaming agent loop
An Express backend drives the Claude API with 213 tools, batching 5 tools per round and streaming responses and reasoning to the React client over SSE.
Financial data service
A typed FMP service layer exposes 230 endpoints across twelve categories, from fundamentals and technicals to SEC filings and alternative data.
Python analysis engine
Eleven Python scientific services run Monte Carlo, Black-Scholes, GARCH, VaR, and portfolio optimization in a managed virtual environment with 17 libraries.
Export pipeline
Reports render to PDF through Puppeteer, to DOCX through Pandoc, and to Beamer slides through LaTeX, alongside CSV/XLSX/JSON data downloads.
Validated persistence
Drizzle ORM over SQLite or PostgreSQL stores sessions, analyses, and token metrics, with Zod schemas validating every boundary.
Tech Stack
Frontend
Backend
AI
Quant & Data
Research & Quality
Highlights
- Live demo at www.vquant.ai — a working, deployed platform, not a prototype.
- Six selectable Claude models with up to 1M tokens of context; the reasoning widget auto-expands while the model thinks.
- 44 documented API endpoints, 6 route modules, 57 React components, and 6 Vitest suites with 41 passing tests across 150+ source files.
- The quant engine spans 8 built-in models, from Monte Carlo simulation to Sharpe, Sortino, Calmar, and max-drawdown risk metrics.
- Four independent web-research providers — Tavily, Firecrawl, Exa, and SerpAPI — give the agent search, scraping, and semantic retrieval.
- Every public analysis in the community showcase exposes the exact tool-chain the agent used, making results auditable.
Explore VQuant
AI financial intelligence platform — the full source is on GitHub.