Jevinik

Jevinik is a stock decision terminal that retrieves live market evidence with Valyu and estimates whether a stock will trade higher in 30 days.

Demo

Watch the 26-second Jevinik demo

Jevinik uses Jev, TypeSafe AI's System One evaluation model, to make typed, probabilistic decisions without calling a general-purpose generative LLM.

The interface reports the probability of a higher price, overall outlook, evidence quality, category-level signals, supporting sources, evidence retrieval time, decision time, and browser-observed completion time.

Features

• Autocomplete for common US and European stocks
• Five Valyu searches executed in parallel
• Fast Valyu web mode for analyst and macro evidence
• Fast, structured Jev evaluation
• Bullish, neutral, and bearish signals by evidence category
• Source-level evidence inspection
• Separate evidence, decision, and end-to-end timing
• Responsive Bloomberg-inspired terminal interface

How It Works

1. The user enters a ticker.
2. The API validates and normalizes the ticker.
3. Valyu searches five evidence categories in parallel:
   • Market data
   • Company news
   • Industry news
   • Analyst views
   • Macro risks
4. Price history is condensed into returns, range, volume, and recent-close statistics.
5. Jev evaluates the compact evidence state using typed boolean, score, and choice questions.
6. The API returns the probability, outlook, evidence quality, category signals, source evidence, and server-side timing breakdown.

SEC filings are currently disabled in the application flow. The Valyu SEC source can be re-enabled later when desired.

Requirements

• Node.js 20.9.0 or newer
• npm
• A Valyu API key
• A Vercel AI Gateway API key

API Keys

Valyu

Valyu supplies the market, news, analyst, and macro evidence.

1. Create or sign in to a Valyu account at platform.valyu.ai.
2. Create an API key in the Valyu dashboard.
3. Confirm the account can access the proprietary stock source used by this app: valyu/valyu-stocks.
4. Add the key to VALYU_API_KEY in .env.local.

The application also uses Valyu news and fast web searches. If Valyu returns Forbidden, check that the key is current, the account has sufficient quota, and proprietary financial-source access is enabled.

Valyu documentation: docs.valyu.ai

Vercel AI Gateway

Vercel AI Gateway routes requests to both typesafe-ai/jev and the configured LLM.

1. Open the AI Gateway section in your Vercel account.
2. Create an AI Gateway API key.
3. Ensure the account can access typesafe-ai/jev and openai/gpt-5.
4. Add the key to AI_GATEWAY_API_KEY in .env.local.

AI Gateway documentation: vercel.com/docs/ai-gateway

Model references:

• Jev on Vercel AI Gateway
• GPT-5 on Vercel AI Gateway
• AI SDK TypeSafe provider

Local Setup

Clone the repository and install dependencies:

git clone git@github.com:unicodeveloper/jevocks.git
cd jevocks
npm install

Create the local environment file:

cp .env.example .env.local

Set the required values:

AI_GATEWAY_API_KEY=your_vercel_ai_gateway_key
VALYU_API_KEY=your_valyu_key

# Optional. Defaults to openai/gpt-5.
LLM_MODEL=openai/gpt-5

Start the development server:

npm run dev

Open http://localhost:3000.

Do not commit .env.local. It is ignored by Git.

Decision Engines

Jev

Jev is not used as a text-generating LLM. TypeSafe AI describes it as a System One evaluation model: a distinct model class built for fast, structured decisions in software. This app calls it through AI SDK experimental_evaluate with model typesafe-ai/jev to answer typed boolean, score, and choice questions against one shared evidence state.

Jev is the default because it is designed to make this structured decision substantially faster than a general-purpose LLM.

GPT-5

The optional LLM path uses AI SDK generateObject with a Zod schema so its response matches the Jev classification contract. It remains available to API clients through decisionEngine: "llm" but is not exposed in the current interface.

The default model is:

openai/gpt-5

Set LLM_MODEL to another AI Gateway model ID to change it without modifying code.

API

POST /api/analyze

Request:

{
  "ticker": "AAPL",
  "decisionEngine": "jev"
}

decisionEngine accepts jev or llm and defaults to jev.

Example:

curl -X POST http://localhost:3000/api/analyze \
  -H "Content-Type: application/json" \
  -d '{"ticker":"AAPL","decisionEngine":"jev"}'

The response contains:

• Normalized ticker
• Selected decision engine
• Evidence grouped by category
• Probability the price is higher in 30 days
• Outlook score and label
• Evidence quality score
• Category-level signals
• Evidence retrieval, decision, and total server timing

Timing

The UI exposes three different measurements:

• Data time: server time spent retrieving and compacting Valyu evidence.
• Decision time: server time spent calling the selected decision engine and mapping its response.
• Complete in: browser-observed time for the entire request, including network transfer and client processing.

API clients comparing Jev with GPT-5 should use decision time. Run the same ticker more than once because provider and network latency vary between requests. The two modes execute separate requests, so their Valyu results can also differ slightly as live sources change.

Scripts

npm run dev    # Start the development server
npm run lint   # Run ESLint
npm run build  # Create and type-check the production build
npm start      # Start the production server

Deployment

For Vercel deployment:

1. Import the GitHub repository into Vercel.
2. Add VALYU_API_KEY and AI_GATEWAY_API_KEY under Project Settings → Environment Variables.
3. Optionally add LLM_MODEL.
4. Deploy.

The analysis route allows up to 60 seconds because GPT-5 can take considerably longer than Jev.

Troubleshooting

VALYU_API_KEY is not set

Create .env.local, add the key, and restart the development server.

Valyu returns Forbidden

• Verify the key in .env.local is current.
• Restart the server after changing environment variables.
• Check Valyu usage and quota.
• Confirm access to valyu/valyu-stocks.

No evidence found for TICKER

Every Valyu category failed or returned no results. Check the server logs for category-specific errors and test the Valyu key directly.

AI Gateway authorization or model errors

• Verify AI_GATEWAY_API_KEY.
• Confirm access to typesafe-ai/jev and the configured LLM_MODEL.
• Check that LLM_MODEL uses the provider/model format.

GPT-5 is much slower than Jev

This is expected for this workflow. Jev performs purpose-built System One evaluation without a general-purpose generative LLM call, while GPT-5 performs structured generation. Compare the API response's decision timings rather than total completion times.

Project Structure

src/
├── app/
│   ├── api/analyze/route.ts  # Validation and analysis endpoint
│   ├── globals.css           # Global terminal theme
│   ├── layout.tsx            # Fonts and metadata
│   └── page.tsx              # Interactive terminal UI
└── lib/
    ├── categories.ts         # Evidence category types and labels
    ├── classify.ts           # Jev and GPT-5 decision engines
    └── evidence.ts           # Valyu evidence collection

Security

• Keep all API keys server-side in environment variables.
• Never expose keys through client components or browser logs.
• Never commit .env.local or other credential files.
• Rotate any key that is accidentally exposed.

Disclaimer

Jevinik is not financial advice. Its estimates are generated from retrieved evidence and model outputs and may be incomplete or incorrect.