score
gate
route
>
route
decide
save tokens
>
JEV
LLMs write. JEV decides. A System One brain that takes state and typed questions and answers with Choice, Score and Noul, in microseconds, with confidence. No chat. No prose. Just the decision.
Most of an agent's calls never needed a writer. They needed a decision.
Is this spam? Which worker goes next? Is the draft ready for a human? Those aren't writing jobs. Sending them to a frontier model is slow, expensive and oddly chatty. Split the loop in three.
If it creates → LLM
Research, drafting, explaining. The expensive model does what it's brilliant at, and only that.
- write the reply
- draft the briefing
- explain the clause
If it picks, scores or gates → JEV
Every fork in the loop becomes a typed question with explicit options. Answers come back as data your code can branch on.
- Choice · research / write / review
- Score · urgency low → critical
- Noul · safe to publish? yes, 0.93
If it does the thing → code
Execution stays deterministic: limits, retries, saves. Irreversible actions keep a human gate.
- max 10 actions per run
- save progress every step
- publish only on approval
Choice. Score. Noul. That's the whole vocabulary.
Three question types cover almost every gate an agent will ever hit. These are the brain's latest real answers, updating live.
Pick one.
Exactly one labelled option, with a confidence. Which folder, which worker, which narrative.
Rank it.
An ordered label: low → critical, bare → complete. Ratings your code can threshold.
Yes or no.
A binary with confidence. Copycat? Safe to run? Escalate? The gate in front of every action.
Every new Solana launch, decided. Right now.
The moment a token is created on pump.fun, JEV gets its state and four typed questions. Narrative (Choice), metadata (Score), copycat (Noul), dev buy (Score). Each answer is timed in microseconds.
Give it state. Get a decision.
Paste a tweet, an announcement, a forum post, an incident report. The brain answers four typed questions about it. Every call is real and timed.
Small on purpose. Measured in public.
The topic head is a naive Bayes model trained at startup on real crypto pages read live by headless browsers, then tested on pages it never saw. The numbers below come from that test, not a slide.
$JEV
A community token for the brain that decides. The cortex, the playground and the API on this site are live and free to use. No promises beyond what you can watch run.
3uM4LH7puk9QmmpHx38J58Gt2rL9wvudwaLCtfaRpumpBefore you ask.
Is this a chatbot?
No. It never writes a sentence. It reads state, answers typed questions and returns data with a confidence. That's the point: the writing stays with the LLM, the forks move to something fast.
Are the decisions on this page real?
Yes. The cortex listens to pump.fun's public feed of new tokens. Each launch is decided on the moment it appears, and the latency shown is measured on the server for that exact call.
Is it financial advice?
No. The live questions are descriptive (what narrative, how complete the metadata is, whether the name copies a recent launch, how big the dev buy was). None of them say buy or sell, and nothing here should be read that way.
Can I use the brain in my own agent?
Yes, while it's in this early form: the API is open, rate limited and free. Send text, get typed answers back.
Is JEV affiliated with TypeSafe?
No. $JEV is an independent community token. The System One idea it celebrates belongs to the people building it; this site runs its own small brain.