Why use Jev here instead of just asking GPT-5 whether to buy?

Jev answers typed boolean, score, and choice questions against one shared evidence state through experimental_evaluate, which keeps the verdict structured. The GPT-5 path uses generateObject with a Zod schema so it matches the same classification contract.

Can I trade directly off the probability it shows?

No. The interface reports higher-price probability, outlook, evidence quality, category signals, sources, and three timings, plus a disclaimer that it is not financial advice. Treat it as a research aid and verify before acting.

What setup is required before the first ticker works?

Create keys for Valyu with valyu-stocks access and Vercel AI Gateway with typesafe-ai/jev access, copy .env.example to .env.local, set VALYU_API_KEY and AI_GATEWAY_API_KEY, then run npm install and npm run dev.

How should I fairly compare Jev decision time against GPT-5?

Use the decision-time field for engine comparisons, run the same ticker more than once, and expect some Valyu drift between runs. Total completion time includes network and client work, so it overstates the model gap.

What evidence does it search, and from where?

Valyu runs five searches in parallel across market data, company news, industry news, analyst views and macro risks, using the proprietary valyu/valyu-stocks source plus news and fast web modes. Price history is condensed into returns, range, volume and recent-close statistics before Jev sees it. SEC filings are currently disabled in the application flow.

What are the three timing numbers it shows?

Data time is server time spent retrieving and compacting Valyu evidence, decision time is server time calling the engine and mapping its response, and complete-in is the browser-observed end-to-end time including network and client work. To compare engines fairly, use decision time.

What runtime does it need, and how do I call the API?

Node.js 20.9.0 or newer plus npm, with Valyu and Vercel AI Gateway keys. POST to /api/analyze with a body like {"ticker":"AAPL","decisionEngine":"jev"}; decisionEngine accepts jev or llm and defaults to jev, and the response carries the normalized ticker, evidence grouped by category, probability, outlook, evidence quality, signals and timings.

CommunityJev Tools & CLIs

Jevocks

Stock decision terminal that gathers Valyu market evidence across five categories and uses Jev for a typed 30-day outlook with probability, signals, and timing.

See highlights

Open github.com

Stars: 36
Language: TypeScript
Published by: Community
Author: unicodeveloper
Last verified: Oct 2026

unicodeveloper/jevocks

User repository on GitHub

View on GitHub

Everyday Stocks Status with Jev

Forks: 10
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

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

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Five Valyu searches run in parallel across market, company, industry, analyst, and macro evidence.
Jev evaluates a compact evidence state with typed boolean, score, and choice questions.
Bloomberg-inspired terminal shows probability, outlook, evidence quality, and source inspection.
Separate data, decision, and browser-observed timings plus a POST /api/analyze endpoint.

Watch out

Requires Valyu and Vercel AI Gateway keys with access to valyu-stocks and typesafe-ai/jev; SEC filings are disabled in the flow and estimates may be incomplete. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-recall

Python memory filter that judges every stored memory against the current request with one calibrated yes/no per memory in a single Jev request, keeping everything above a threshold.