User repository on GitHub
Crypto trading bot on Binance testnet using TypeSafe (Jev) to judge news
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Tools & CLIs
Trading Bot Jev
Medium-term crypto trend-following bot that uses Jev to read and sort crypto, political, central bank, regulator and exchange news, then applies the result as a small news bias and a regulatory veto on a 4-hour EMA strategy.
- Stars
- 3
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- Spykoninho
- Last verified
- Oct 2026
About this project
Un bot de trading crypto construit pour répondre à une question : à quoi sert Jev, le modèle « System One » de TypeSafe, et apporte-t-il quelque chose par rapport à un algorithme seul ?
Quoted from the project's README. Read the full README.
- Languages
- TypeScript
Highlights
- Jev reads, the algorithm decides: typed judgments like a 0.97 probability shift the two thresholds by at most 0.5%, and it never predicts a price.
- One hard-coded event rule - a political post judged pro-crypto at 0.8 or above buys 10% of equity and sells four hours later - resting on 13 events.
- Paper portfolio by default with real prices, simulated fees and no exchange key; --real mirrors paper orders live and leaves the rest of the account alone.
- npm run backtest compares no-news, bias-only, veto-only, bias plus veto and buy-and-hold, with -- --fees replaying at several fee levels.
Watch out
The result to read first is a negative one: the author measured the news effect three times and got -2.8, -2.1 and +4.9 points depending on currency and fees, so the sign changes and it is treated as noise. The README is equally blunt that the event rule is a hypothesis with n=13, not a result, and that the event study found the price had already moved before 1,201 high-impact headlines published, leaving +0.032% at one hour, far under fees. The limits section lists one flat measurement period starting after FTX, selection bias from choosing SOL and dropping AVAX on that same period, three assets correlated around 0.8, short history for the stablecoin pairs, dollar-denominated returns, issuer and depeg risk on the stablecoin, and an order book about 40 times thinner than the euro book, hence SLIPPAGE_BPS=4. Possible contamination is acknowledged because the model is trained after the fact, the political posts arrive through unofficial third-party aggregators, and the pre-1.0 SDK is a single vendor. The bot also refuses to start without a TypeSafe key even in paper mode. The README continues past the point this listing's source text ends, cut off inside the setup walkthrough. 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.
The README answers directly: Jev is not measurably useful on prices, with three measurements giving -2.8, -2.1 and +4.9 points depending on currency and fees, so the sign changes and it is noise. What it buys is throughput - it read and sorted 49,496 texts for roughly 2.50 in API cost, where a conventional LLM would be slower and more expensive. The bot also keeps trading on the trend rule alone if the TypeSafe API goes down, since the pre-1.0 SDK is a single point of failure.
Not micro-trading: a medium-term trend follower on 4-hour candles, an EMA over 200 candles covering about 33 days, buying when a candle closes above EMA x 1.01, selling everything below EMA x 0.99 and waiting in between, across BTC, ETH and SOL with a 10% cash reserve. By default nothing real is at stake - npm start keeps a paper portfolio in data/ with live prices, simulated fees and slippage, and sends no orders anywhere. Execution is on Bitvavo in the stablecoin, and the README states plainly that nothing here is investment advice.
Real mode takes a flag you cannot reach by accident: npm start -- --real, and the paper portfolio stays the brain, mirroring each paper order to the market. It records the exact quantity received at purchase and only sells that, reconciles the balance at startup and hourly with a banner on a mismatch, and refuses to send an order on a price over 30 seconds old. The README recommends rehearsing with a very small START_CASH, 5 per order and 10 on the paper side, above hard floors. One honest limit: the exchange does not expose per-key permissions via API, so the bot cannot check your key scopes.
It is deliberate. The README puts it at 14 to 16 round trips per asset per year, because each one costs fees twice plus slippage and the gain comes from a few long trends held to the end. The measured profile is about a quarter of trades winning, a negative median of -2.41%, and the five best trades making 206% of the cumulative gain. Losing three times out of four and still coming out ahead is normal for trend following, but the README is candid that it takes holding the losses.
Only two: a news bias that shifts the two trend thresholds by up to plus or minus 0.5%, and a regulatory veto on buys. Jev judges each news item for the asset concerned, sentiment, materiality and regulatory risk in about 100 ms, and the algorithm computes the trend, sets thresholds, sizes and executes.
On 4-hour candles, a close above EMA 200 times 1.01 buys, a close below times 0.99 sells everything, and in between it waits. The EMA over 200 candles covers roughly 33 days. Capital is split equally across the chosen assets minus a 10% cash reserve.
The bot retains the exact quantity received on a buy and resells only that amount, reconciles the balance at startup and hourly with a red banner on any mismatch, and refuses to send an order on a price older than 30 seconds. It should run on a dedicated account with a trade-only API key that has no withdrawal rights.
More Jev resources like this
3GitHub stars
A CLI that sends state plus typed questions to Jev and prints structured JSON, with boolean, choice and score question types, a documented exit-code contract for CI, and an optional Vercel AI Gateway transport.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
jtsang4/jev-cliTypeScript
jtsang4/jev-cliTypeScript0GitHub stars
Shared 1,000 by 1,000 emoji canvas where Jev paints alongside visitors: after each stroke it picks a keyword-matching emoji and a named position relative to that stroke, and answers whether the stroke is unfinished.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
cwdx/1-million-emojisTypeScript
cwdx/1-million-emojisTypeScript197GitHub stars
A Chinese-first desktop workbench for local models whose JEV page runs typed noul, choice and score questions against a local engine or a cloud key, alongside a gateway, RAG knowledge base and shared memory.
Jev Tools & CLIs#community#jevbest
Community
kunpengtalk/OmniStudioTypeScript
kunpengtalk/OmniStudioTypeScript38GitHub stars
Research dashboard for Polymarket BTC five-minute Up/Down rounds that feeds market state to Jev class scores and paper-trades the edge.
Jev Tools & CLIs#community#jevbest#catalog-import
46GitHub stars
Desktop and web live-call coach that sends transcript lines to TypeSafe Jev after each sentence and shows a 0-100 gauge with a suggested next step.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
ZeroGold/call-coach-aiHTML
ZeroGold/call-coach-aiHTML38GitHub stars
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.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
samdotmak/jev-recallTypeScript
samdotmak/jev-recallTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
The request shape barely changes across surfaces; the surrounding contract does. When to call the API directly and when a gateway, a Worker binding, or a router entry earns its extra hop.
Jev Tools & CLIs#gateway#vercel#cloudflare
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Jev labels a record that already exists. Code moves, flags, or holds it. How the document, mail, and transfer tools keep the action out of the model.
Jev Tools & CLIs#classification#email#security
Read article
From the Jev community
Posts from builders shipping with Jev right now.