Stock Research Engine

Multi-agent stock research app that produces executive research memos (≤2 pages) from a single "Research Ask" prompt. Three specialist agents run in parallel (with web search), then a Chairman model synthesizes a consensus Markdown memo.

Stack

• Next.js 14 (App Router), Tailwind CSS
• OpenRouter for LLM calls (Claude, GPT, Gemini)
• Tavily for web search used by each agent

Deploy on Vercel (recommended — no “dev server” needed)

Memo history (Google Drive)

Optional: Sign in with Google to save memos to your Google Drive. Memos are stored in a folder you choose (set GOOGLE_DRIVE_MEMOS_FOLDER_ID in Vercel). Only you can see or delete them. See GOOGLE_DRIVE_SETUP.md for OAuth and folder setup. No secrets go in the repo.

---

Deploy on Vercel

Want a real website URL? Use the full walkthrough: SETUP_GITHUB_AND_VERCEL.md — step-by-step instructions to put the project on GitHub, then deploy on Vercel (written for non-engineers). You’ll get a link like https://your-app.vercel.app to use in your browser.

---

Local setup (optional)

1. Copy env and add keys:

   cp .env.example .env.local

   Edit .env.local:

   • OPENROUTER_API_KEY — from OpenRouter
   • TAVILY_API_KEY — from Tavily

2. Optional model overrides (in .env.local):

   • OPENROUTER_MODEL_ALPHA — e.g. anthropic/claude-opus-4.6
   • OPENROUTER_MODEL_BETA — e.g. openai/gpt-5.2
   • OPENROUTER_MODEL_GAMMA — e.g. google/gemini-3-pro
   • OPENROUTER_MODEL_CHAIRMAN — e.g. anthropic/claude-sonnet-4

3. Install and run:

   npm install
   npm run dev

   Open http://localhost:3000.

Flow

1. User enters a Research Ask (e.g. ticker, time frame, focus).
2. One Tavily search runs for that ask (time frame + technical/context).
3. Agent Alpha (Claude), Agent Beta (GPT), Agent Gamma (Gemini) run in parallel, each with the same search context. Each produces:
   • Data analysis aligned to the ask
   • Technical quant-style analysis (trend levels, stochastics, options volume, institutional flow)
   • JSON with citations and confidence.
4. Chairman model reads all three JSON reports and writes a single Markdown executive memo (consensus, key levels, risks, bottom line; ≤2 pages).
5. Frontend shows the memo rendered as Markdown.

What I need from you

• API keys: OPENROUTER_API_KEY and TAVILY_API_KEY in .env.local.
• Exact model IDs (if you want specific models): set OPENROUTER_MODEL_ALPHA, OPENROUTER_MODEL_BETA, OPENROUTER_MODEL_GAMMA to the IDs from OpenRouter Models (e.g. Claude Opus 4.6, GPT-5.2, Gemini 3 Pro when listed).
• Vercel (optional): for production, set the same env vars in the project settings and optionally maxDuration in app/api/research/route.ts to match your plan (e.g. 120s).

Once keys and (optionally) model IDs are set, you can run the app and generate memos from the Research Ask input.