Stock Indicator Dailies

An AI-driven technical-analysis utility that automates the daily ritual of checking market signals.

Stock Indicator Dailies turns the repetitive chore of logging into a charting platform, configuring indicators, and eyeballing crossovers into a single click. You enter a ticker; an autonomous browser agent logs in, applies a fixed indicator set, screenshots the chart, and a Visual Language Model (VLM) reads it back to you as a Buy / Sell / Hold signal.

[!IMPORTANT] This is not financial advice. Stock Indicator Dailies is a data-acquisition and reporting tool. Every final investment decision remains the sole responsibility of the user.

---

How it works

  User enters ticker (e.g. NVDA)
            │
            ▼
  ┌───────────────────┐   logs in, applies indicators, screenshots
  │  Agentic Layer    │──────────────────────────────────────────────┐
  │  (Playwright)     │                                               │
  └───────────────────┘                                               ▼
            │                                              high-res chart PNG
            │                                                         │
            ▼                                                         ▼
  ┌───────────────────┐   interprets crossovers vs. signal   ┌───────────────────┐
  │  VLM Analysis     │◀─────────────────────────────────────│  Chart image      │
  │  (Claude)         │   criteria → structured JSON          └───────────────────┘
  └───────────────────┘
            │
            ▼
  Daily Report card: screenshot + Buy / Sell / Hold recommendation

Signal criteria

The signal is derived from three indicators with fixed chart parameters, but a tunable decision policy; the consensus thresholds and recency window are all adjustable from the web UI's Indicator Settings panel (defaults shown below).

Indicator	Buy signal	Sell signal
MACD (8, 17, 9)	Bullish crossover (MACD line crosses above signal) below zero	Bearish crossover (MACD line crosses below signal) above zero
Slow Stoch (14, 5, 3)	%K crosses above %D while oversold (< 20)	%K crosses below %D while overbought (> 80)
10-day SMA	Price crosses above the 10-day SMA while it's sloping up	Price crosses below the 10-day SMA while it's sloping down

A crossover only counts if it happened within the last 3 days (default); older ones read as no signal. The three per-indicator reads are then combined: SELL requires unanimity (3 of 3), BUY requires only 2 of 3, otherwise HOLD; backtesting showed this asymmetric policy (easy to exit, harder to re-enter) outperforms a symmetric one more consistently. See evals/backtest and the in-app Historical Testing panel to validate any policy change against real history before trusting it.

Tech stack

Layer	Choice	Purpose
Frontend	Next.js (Vercel)	Dashboard, Indicator Settings, Historical Testing UI
API	Hono (Render)	Pipeline orchestration, backtest, AI-advisor endpoints
Agentic layer	Playwright	Log in, configure indicators, capture chart
AI layer	Claude (vision read + web_search tool-use advisor)	Chart interpretation; settings-tuning research
Database	Supabase	Chart cache, AI-suggestion cache, failure logs
Evals	Labeled ground truth + walk-forward backtest	Guard against interpretation drift & bad policy changes

Repository layout

stock-indicator-dailies/
├── apps/
│   ├── web/             # Next.js dashboard (frontend, deployed on Vercel)
│   └── api/              # Hono API: pipeline orchestration, caching, job/poll routes (deployed on Render)
├── packages/
│   ├── agent/            # Playwright browser-automation layer
│   ├── vlm/              # Claude vision analysis + prompts
│   ├── advisor/          # Claude + web_search settings-tuning advisor
│   ├── daily/             # Daily pipeline orchestration (capture -> analyze -> report)
│   ├── indicators/        # Deterministic indicator math (MACD/Stochastic/SMA/ATR/ADX)
│   └── shared/            # Shared types & signal-criteria definitions
├── evals/
│   ├── interpretation/   # Buy/Sell/Hold vs. labeled ground truth
│   ├── backtest/          # Walk-forward backtest of the policy vs. buy-and-hold
│   └── retrieval/         # SSIM-based chart-retrieval eval (target >= 95%); not started
├── supabase/              # Schema & migrations
└── docs/                  # PRD, architecture, roadmap

Getting started

Prerequisites: Node.js 22+, npm, and a Playwright-capable environment. AI integration requires a Claude API key; the chart and AI-suggestion caches require a Supabase project (see .env.example for the full list).

git clone https://github.com/JTM-Kaihatsu/stock-indicator-dailies.git
cd stock-indicator-dailies
cp .env.example .env.local   # then fill in your keys
npm install

npm run dev:api    # Hono API on :3001
npm run dev:web    # Next.js dashboard on :3000, in a second terminal

npm test           # full test suite across every package
npm run typecheck  # every workspace

Security & privacy

Because the agent authenticates into external charting/brokerage platforms on your behalf, security is a first-class concern. See docs/ARCHITECTURE.md for the full model. Highlights:

• Read-only, no trade execution; trade execution is explicitly out of scope for the MVP.
• Encrypted, local-first credentials; secrets live in OS-level vaults (e.g. Keychain), never in plaintext or logs, never on centralized servers.
• Token-based auth preferred; OAuth / scoped read-only API keys over passwords.
• Image sanitization; screenshots are regionally cropped to strip account-identifying metadata.

Roadmap

Phase II extends Stock Indicator Dailies into financial-statement extraction and analysis. See docs/ROADMAP.md.

License

No license chosen yet. This public repository is currently "all rights reserved" by default. Add a LICENSE file to grant usage rights.