Why does the voice differ from plain DeepSeek output?

Same endpoint and prompt, different chooser. Greedy and temperature sampling both score near the top of the AI detector while Jev-chosen replies score near the bottom, because the token chooser changed rather than the temperature.

Does it stay coherent for long essays?

Jev on every token runs out of things to classify after roughly a paragraph and dissolves. The essay preset stretches coherence with gating and stop wording, but detector scores still flag long-form output.

What keys and setup does the REPL need?

Needs a DeepSeek key plus an OpenRouter key serving Jev on the System One endpoint, with an optional Pangram key for detector scoring. Install via pip from git and copy the example env file.

How do you stop the grammar glitches and repeats?

Gated presets skip Jev when DeepSeek is highly sure so it carries grammar while Jev takes branch points, plus candidate floor, order averaging, repeat penalty, n-gram block, and patience-counted ramble verdicts.

What do the four presets change?

gated (default) lets DeepSeek carry grammar while Jev takes branch points, using weight (1 - p_top)^0.5 and skipping Jev when DeepSeek is at least 99.7% sure; chat gives Jev every token for the strongest, least stable voice; essay applies gating and essay-worded stop questions for long form; pure is the original experiment with no floor and no stop questions.

How does it get a stop token from a model that never chooses one?

Stopping is not a token, so Jev's own is-it-finished and has-it-started-rambling noul judgements ride the same call, with thresholds --stop-noul 0.7 and --ramble-noul 0.6. Because DeepSeek returns no logprobs when its sampled token is EOS, jevseek re-asks at T=2 and un-tempers so the end token becomes a candidate with a true probability.

How much does a token cost, and what must I install?

About one second and $0.00005 per token, versus jevgpt's six calls a word over a 20,000-word dictionary. Install with pip from git and copy .env.example, then supply a DeepSeek key and an OpenRouter key, where Jev is served as typesafe/jev-1.13 on the System One endpoint; a Pangram key is optional for detector scoring.

CommunityJev Tools & CLIs

Jevseek

Experiment turning Jev into a sampler: DeepSeek proposes top-k next tokens with logprobs and Jev picks each token, also judging when to stop or cut rambling.

See highlights

Open github.com

Stars: 4
Language: Python
License: MIT
Published by: Community
Author: blingdivinity
Last verified: Oct 2026

blingdivinity/jevseek

User repository on GitHub

View on GitHub

DeepSeek proposes the next token, TypeSafe's Jev chooses it: a decision model used as a sampler

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

DeepSeek proposes the next token. Jev chooses it. TypeSafe's jev is a decision model: it cannot generate text, it can only answer typed questions about a piece of state — pick one of these, yes or no, score this. jevseek makes it a sampler. Each step, deepseek-flash returns its top-k next tokens with logprobs; jev is shown the conversation and asked "Which token should come next?"; the token it picks is appended; repeat. Jev also decides when the reply is finished, and when it has started rambling.

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

Tagscommunity awesomejev catalog-import

Works forJev for Classification Jev for Routing

Used inJev for Software & DevTools

Languages: Python

Highlights

Each step shows DeepSeek candidates to Jev as which-token-next, appending the pick until stop or ramble verdicts fire.
Gated, chat, essay, and pure presets trade Jev control against stability, all overridable via CLI flags.
Verbose mode shows per-token offers versus picks; record mode saves detector score plus full per-step trace.
Measurements and recorded runs with every candidate and judgment ship in FINDINGS and rendered run pages.

Watch out

Experimental sampler where long outputs can dissolve, repeat, or ramble; distinctive voice does not imply undetected text. 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

jevclip

Video tool where Jev scores subtitle segments and code assembles a short highlights cut plus a de-watered full cut with timestamped summaries.

Jev Tools & CLIs#community#jevbest#catalog-import

Community

cclank/jevclipPython

jev-resume-disqualifier

Recruiter dashboard that applies calendar date math plus visa, tenure, license and clearance gates, then a Jev decision step, sorting resumes into disqualified, advance or manual review.

Jev Tools & CLIs#community#jevbest#catalog-import

Community

AiPersonacademy/jev-resume-disqualifierPython

jev-trade-cc

A point-in-time backtest of a William O'Neil momentum strategy where jev picks the entries and decides the exits, paired against the same rules with no model — and honest enough to report the model as indistinguishable from them.

Jev Tools & CLIs#community#awesomejev#catalog-import

Community

michaelpersonal/jev-trade-ccPython

Jev-Register-Tool

Pure-HTTP automation for the TypeSafe register, magic-link login, onboarding, and API-key creation flow, with pluggable temp-mail backends, concurrent batch runs, and key verification exports.

Jev Tools & CLIs#community#awesomejev#catalog-import

Community

2951461586/Jev-Register-ToolPython

dohnuts

Python toolkit plus 0.8B multimodal model that takes a message, document, or image and returns calibrated choice, yes/no, and score answers in one forward pass for local decision workloads.

Jev Tools & CLIs#community#awesomejev#catalog-import

Community

PsiACE/dohnutsPython

JEV-Paper-Radar

Forkable weekday radar that asks Jev one yes-or-no question per interest over each new arXiv announcement and publishes a page plus RSS feed.

Jev Tools & CLIs#community#jevbest#catalog-import

Community

Eliot5566/JEV-Paper-RadarPython

From the Jev guides

Original write-ups on what this entry is for.

Jev on the Gateways: Vercel, Workers AI, and OpenRouter

Oct 12, 2026

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

Exporting Decision Models: GGUF, ONNX, and Device Runtimes

Oct 10, 2026

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

Records at Rest: Documents, Inboxes, and Risk Queues

Sep 28, 2026

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

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai