54%
of retail investors have used an AI chatbot for investing research
Investing.com, retail AI survey, 2026
The company behind VYNN
Research for
the rest of us.
We believe the quality of your research should never depend on the size of your portfolio.
- investors in early access
- 5,000+
- for a full analysis
- ~2 min
- research tools
- 20
Our conviction
Wall Street's last data moat.
Hedge funds pay $24,000 per Bloomberg seat each year for the research that moves markets. You get free chat rooms and 15-minute-delayed quotes. The AI tools built to close that gap make up their numbers.
We want to democratize financial analysis for everyone. VYNN runs the institutional pipeline for anyone with a phone: fundamentals, news intelligence, a sourced valuation, and a rating with its confidence stated. It answers in your language, it shows its inputs, and it flags the gap when its value and the analysts' are far apart.
Why this moment
Everyone can ask. The answer must hold up.
The demand for a verdict is old: most investors read an analyst's opinion, not the filing. What is new is that the verdict now comes from a model that invents the number. VYNN exists for the gap between those two facts.
31%
trust AI for financial advice
Investing.com, 2026
71%
consult analyst opinions before they trade; almost none read the filings
Laarits & Wurgler, NBER WP 33625
What we’re building
Power, with guardrails.
A full research workflow, from the data to the model to the report, and a record you can return to. Built around three commitments.
Power
A stock in any language, an ETF, a coin, an option, a portfolio, or the market's odds on a Fed cut. VYNN reads what you need and calls only the tools that answer it. No menus, no fixed pipeline.
20 tools
four asset classes, any language
Guardrails
Every figure is computed by deterministic code with its source printed beside it. In the report, a sentence that quotes a number the model didn't produce is removed. When VYNN's value and the analysts' are far apart, it flags the gap and shows both.
≥95%
of a recommendation's sentences must cite a source
Speed
A human analyst needs six to twelve hours per stock. VYNN needs about two minutes for a full report, a DCF workbook and a rating it is allowed to publish. The stages run in parallel, which cut the original pipeline by 78.6%.
78.6%
latency cut by parallel stages
VYNN in numbers
Open any figure to see how it is measured.
~2 minFull analysis, end to end
Financials, a DCF workbook, news screening and a written report. Recorded runs took 2 min 12 s (Microsoft, 29 Sep 2026) and 2 min 34 s (Costco, 2 Oct 2026). Quick questions come back in seconds because the agent runs only what each question needs.
78.6%Latency cut by parallel stages
Measured against the original sequential pipeline in August 2026. The model and the news screening run at the same time, and the report's sections are written in parallel.
≥95%Recommendation sentences that must cite a source
A check in code requires at least 95% of a recommendation's sentences to cite a source. A draft that fails is rewritten, and one that still fails carries a validation warning. Separately, a report sentence that quotes a number the model didn't produce is removed.
5KBeta users
Live at app.vynnai.com and free during early access.
20Research tools
Prices, technicals, news, macro data, financials, the DCF model, reports, comparisons, crypto, funds, options, portfolio risk and optimization, prediction markets, and a live chart in the chat. VYNN calls only the ones a question needs.
100Companies in the nightly regression
Every night, valuations run against a golden dataset of 100 QQQ companies. A release is blocked if any valuation drifts beyond its threshold.
~$0.03Model cost of a full report
On the model VYNN runs today. A full Apple report measured $0.06 on 12 Sep 2026, on a model that cost more per token. A quick answer costs a fraction of a cent.
4Asset classes
Companies, funds, digital assets and prediction markets, each with its own research workspace.
A note from the founder
The number.
The arithmetic.
The evidence.
I started VYNN because the research that moves markets is sold by the seat, and the AI tools built to replace it make up their numbers. A retail investor deserves the same thing an analyst gets: the number, the arithmetic, and where each input came from.
VYNN is live and free, with 5,000 beta users. Every figure is computed in code, every input is printed with its source, and when VYNN's value and the analysts' are far apart, it flags the gap and shows both.
Zanwen Fu
MS CS, Duke · Previously Robinhood, Binance, and AutoCodeRover (acquired by Sonar).
Under the surface
Read the code. Open source.
VYNN is three repositories under the Agentic-Analyst organization. The agent and its valuation engine are public; the API and the web app are private for now. Read the agent loop, the DCF engine and the cost-of-capital feeds yourself.
stock-analystPublic
The agent backend. A tool-use agent reads a request in any language and picks the tools it needs; four analysis sub-agents run as tools. A 10-tab DCF engine with the cost of capital built from published data, 34 prompt templates, and a nightly valuation regression across 100 companies.
Python · Agent · LangGraph
api-runnerPrivate · opening soon
FastAPI orchestration. Job lifecycle, SSE streaming logs, WebSocket price feed, an ephemeral Docker container per analysis for isolation, MongoDB persistence, and Redis rate-limiting.
Python · FastAPI
vynnai-webPrivate · opening soon
The frontend. React 18 + Vite, shadcn/ui, Recharts, real-time SSE + WebSocket, multi-provider OAuth, and the research workspaces for companies, funds, digital assets and prediction markets.
TypeScript · React
Let’s talk
Believe in what we’re building?
For partnerships, investment conversations, or a question about VYNN, reach the founder directly.
zanwen.fu@duke.edu