Why use Jev instead of an LLM agent for the screening?

The author measured $1.1-$1.5 a morning versus $8-$15 for an Opus agent, because Jev answers fixed-choice questions without re-reading a growing agent context on every turn.

What happens when Jev is unsure about a new paper?

Borderline items go to the note's Review list for the human to decide, rather than being dropped or accepted silently; ticks on kept papers also seed the next run.

What does a morning run actually cover?

It takes the next three topics in turn, a setting you control, checks arXiv, Hugging Face, Semantic Scholar, OpenAlex and GitHub against two to four open questions per topic, and writes one Markdown note per topic into your Obsidian vault. The author has run it every morning since 2026-09-24 across seven topics.

How are the citations in a note produced?

Each [n] points to a sentence copied word for word from its source, never a paraphrase, so any citation can be checked against the original. An [Inference] paragraph at the end is jrp's own output rather than a quote. In the Chinese and Japanese notes third-party sentences are replaced by links instead.

How do I install and schedule it?

Python 3.12 and uv, then uv tool install jrp and jrp init --lang en, which writes ~/.config/jrp with env, config.toml and a questions folder. Fill in JRP_VAULT_DIR and TYPESAFE_API_KEY, pick a writer with JRP_PROSE_MODEL, then jrp doctor checks setup without running a job and jrp try --line writes a throwaway note. jrp schedule install --hour 5 writes a launchd job on macOS.

What can I control, and what is the hard dependency?

You write the questions: the question itself, what it is not about, the search queries per source, and a paper it must always keep, whose absence the note flags as screening drift. jrp questions new --line opens a Claude Code session to draft them, and no model writes a search query at run time, which is how the author avoided the old agent drifting onto one theme. Jev is not swappable: none of the four local models tried reproduced its decisions.

How does it stop the run from drifting onto one topic?

No model writes a search query at run time; code fixes which sources are searched and how often, and every note counts how many distinct subject areas its kept papers cover so drift shows up as a falling number. The author's previous Opus agent chose 88 of 238 topics, 37%, on a single theme in its first three months.

CommunityJev Tools & CLIs

Jev Research Pipeline

A CLI that checks new papers, repositories and feeds each morning against your open research questions, asking Jev Noul gates and Score dimensions through Pydantic AI's typesafe model; an LLM writes only the prose into Obsidian notes.

See highlights

Open github.com

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

shimo4228/jev-research-pipeline

User repository on GitHub

View on GitHub

A command-line tool that writes a research note each morning on your open questions at about a tenth of an LLM agent's cost: Python runs the loop, TypeSafe Jev screens new papers, and an LLM writes only the prose (English, Chinese or Japanese). Pilot.

Forks: 1
Last pushed: Updated Oct 2026

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

About this project

A research note each morning on the questions you follow, at about a tenth of what an LLM agent cost me: a judgment-only model screens the new papers, and an LLM writes only the prose. A sample note · Why jrp · Try it · Status and limits.

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

Tagscommunity research cli pydantic-ai

Works forJev for Classification Jev for Decision gating Jev for Document processing

Used inJev for Education

Languages: Python
Platforms: CLI macOS
Capabilities: Open sourceRequires API key

Highlights

Run every morning since 2026-09-24 on seven topics; the author measured $1.1-$1.5 a morning, of which $0.03-$0.17 is Jev.
Sources include arXiv, Hugging Face, Semantic Scholar, OpenAlex and GitHub; notes land in an Obsidian vault.
Offline tests replay recorded cassettes; jrp doctor checks every requirement without running a job.
Writer chosen via JRP_PROSE_MODEL (OpenAI Codex, Claude Code or DashScope Qwen); notes in English, Chinese or Japanese.

Watch out

A self-described pilot with hand-tuned thresholds and no fallback judge if TypeSafe changes Jev's price or availability. MIT licensed.

Questions about this entry

Short answers drawn from the linked project or page.

Related terms

Glossary definitions related to this entry.

Autoresearch

More Jev resources like this

ask-jev

Zero-dependency Python CLI that routes small semantic judgments to Jev for agents and CLI pipelines: choose (Choice), check (Noul), score (ordered ratings), batch (1-32 questions in one request) and purify (verbatim passage extraction). GPL-3.0 licensed.

Jev Tools & CLIs#community#cli#decision-routing

Community

logicrw/ask-jevPython

3.4k

Vercel Labs' Unix-like coding agent CLI written in Zig, shipping a typesafe_permission_reviewer builtin that runs the agent's permission decisions through Jev rather than an LLM call. Model-agnostic and embeddable via libfx and ACP.

Jev Tools & CLIs#community#coding-agents#cli

Community

vercel-labs/fxZig

DocJev

523

Classifies and splits PDFs, DOCX, and PPTX with local page text from LiteParse and category decisions from Jev.

Jev Tools & CLIs#community#python#document

Community

jerryjliu/docjevPython

Edward

Supervisor for unattended coding agents: it tracks cross-turn state and asks one batched Jev Choice whether a trajectory should continue, pause or escalate, while deterministic code owns dangerous-command blocking, budget caps and Ed25519-signed receipts.

Jev Tools & CLIs#community#guardrails#safety

Community

VeridicalTech/EdwardPython

GUI JEV Harness

Python harness for GUI grounding: it divides a screenshot into grid tiles, has a vision model describe each tile neutrally, then asks typesafe-ai/jev to choose among the descriptions. Local confidence and margin gates decide whether to descend or refuse, and it emits only a point and bounding box.

Jev Tools & CLIs#community#gui-grounding#computer-use

Community

ZihuaEvan/GUI_JEVPython

jeff

Go CLI for Jev with noul, choice, score, eval, rank, and models commands: rank sends one Score question per item per weighted YAML dimension in a single request and sums weight times score in code, while threshold flags turn a failed gate into exit code 10 for scripts and CI.

Jev Tools & CLIs#community#cli#go

Community

saembit/jeff-cliGo

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