Does the model place trades by itself?

No. The model only answers questions such as direction over three minutes, whether the move is over, setup cleanliness, trend quality, repeat-failure risk, reversal, and thesis intact. Stops, targets, and trailing stops run in code on every tick.

How much will polling every second cost me?

Each agent shows calls, tokens, and spend, and a new model check happens only after the price moves by a set fraction or a set interval passes. On a calm tape most checks reuse the prior answer and cost nothing.

What setup differs between stocks and crypto?

Stocks need an Alpaca key for live prices, trade the regular session only, and close out late in the day with cash-like settlement. Crypto uses the public Coinbase feed with no key and trades around the clock.

What stops it from churning on noise?

A position cannot close in its first 45 seconds and an exit needs two consecutive checks. Thresholds for buying, closing, chasing, and reversal are fields in the agent settings, and every check is logged with latency and cost.

Can it connect to a broker or place real orders?

No. It is paper-only: it simulates fills at real prices and never places an order anywhere, with no exchange, wallet, or broker. Nothing it does or says is investment advice, and the stock data, crypto data, and model answers all come under their own providers' terms.

What does the model actually see for each check?

One code-built description of the market, as plain text facts: the last trade and its age, best bid and ask, moves over ten seconds, one minute, and five minutes, buying and selling pressure, candle shape, volume against normal, opening range, daily trend, RSI, the position with its stop and target, the book, and every trade it already closed there today.

Which operating systems and Node versions does it need?

A desktop app for Windows and macOS built on Electron. It needs Node.js 22 for best results, with 20.19 or newer working, plus Git; run npm install then npm run dev. You can build installers with npm run dist:win or npm run dist:mac.

CommunityJev Practices & Patterns

Jev Realtime Trading

Desktop paper-trading app for stocks and crypto where Jev answers narrow market questions each second and code turns the answers into orders under your stops and targets.

See highlights

Open github.com

Stars: 7
Language: TypeScript
License: MIT
Published by: Community
Author: rthomas24
Last verified: Oct 2026

rthomas24/jev-realtime-trading

User repository on GitHub

View on GitHub

Desktop app for paper-trading stocks and crypto on live prices, with TypeSafe's Jev making the calls and your stops, targets and limits enforced in code. Windows and macOS; never touches real money.

Forks: 4
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Paper-trading agents that watch a live market and decide every second. Pick a stock or a coin, give it some paper money, and watch an AI model read the tape and call up, down or flat, while the app enforces your stops and targets and books every trade. The model is TypeSafe's Jev. It's a desktop app for Windows and macOS. It never touches real money. Screenshots are from a simulated session, for illustration. They are not real or back-tested results.

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

Tagscommunity awesomejev catalog-import

Works forJev for Classification Jev for Tool calling

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Asks direction, chase, setup, trend, repeat-entry, reversal, and thesis questions together and gates orders on thresholds.
Enforces stops, targets, and trailing stops in code on every price tick, not through model judgment.
Caches calm-tape answers until price moves or a timeout passes, with per-agent token and spend display.
Shows verdict bars with thresholds, ordered check history, watchlist tiles, and a grid view of all agents.

Watch out

Paper trading only with simulated fills at real prices; it never places orders and nothing shown is investment advice. 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-agent-authorization

Starter kit putting Kinde identity plus Jev judgment in front of every agent tool call, with allow, hold-for-approval, or deny decided in code and written to a ledger first.