Does fewer tool calls mean more tasks solved?

It cuts tool calls about 20% overall and moves the first gold-file hit to rank 1 in most runs, but resolve rate stayed flat at 15 vs 14 and 4 vs 4. It shortens the path, not the fix.

Is the gain just the briefing format, not Jev?

A control arm with BM25 ranking instead of Jev lands between plain Pi and PiJev, so the gain comes from ranking quality. BM25 top-100 recall at 1 was 0.25 versus 0.74 with Jev reranking.

What does Jev add to per-task cost and wall time?

About $0.011 to $0.014 per agent task in the reported runs, mostly from scoring 100 candidates. A shortlist ablation keeps every rank-1 result at 50 candidates for roughly half the tokens.

What breaks if Jev credentials or services fail?

Node.js 22.19+, npm, and ripgrep. Suggestions stay advisory, and missing credentials, malformed answers, or timeouts fall back to normal coding and lexical search.

Does Jev ever replace compilation, tests, or task acceptance?

No. Jev scores never stand in for compilation, tests, or task acceptance, and its suggestions stay advisory. Missing credentials, service failures, malformed answers or timeouts leave normal coding and lexical search available, and it performs no automatic retry or permission escalation.

What are Jev's three jobs inside the coding loop?

It shortlists skills and checks their instructions for relevance, ranks actual file excerpts retrieved by pijev_search, and classifies a failed tool call to suggest what to check next. The coding model still handles reasoning, edits and tool use, and Jev's suggestions stay advisory.

What did the private-monorepo evidence show, and what is the shortlist default now?

Over 24 merged pull requests in a private TypeScript monorepo, Jev ranked the first PR-edited file higher than BM25 on all 24 and lower on none, moving the median rank from 11 to 1. A shortlist ablation showed 50 candidates keep every rank-1 result of 100 at half the tokens, so the default is now 50.

CommunityJev Agents & MCP

Pijev — tonyzdev

Terminal coding agent on Pi where Jev shortlists skills, ranks retrieved excerpts, and triages tool failures while the coding model handles reasoning and edits.

See highlights

Open github.com

Stars: 6
Language: TypeScript
License: MIT
Published by: Community
Author: tonyzdev
Last verified: Oct 2026

tonyzdev/pijev

User repository on GitHub

View on GitHub

PiJev: a terminal coding agent with Jev in the loop — Jev ranks the repository's files before the first call, picks skills and triages failures; your coding model writes the code. Built on Pi.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

This is one of several Jev listings for the same project. The canonical entry is Pijev — TypeLLM.

About this project

A coding agent with Jev in the loop. PiJev is a terminal coding agent built on Pi, with Jev helping select skills, rank code excerpts, and diagnose tool failures. Your coding model handles reasoning, edits, and tool use. Jev evaluates small, typed questions along the way. Jev for decisions. Your coding model for code.

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

Tagscommunity awesomejev catalog-import

Works forJev for Classification Jev for Routing Jev for Ranking Jev for Tool calling

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Jev shortlists skills, ranks file excerpts with paths and line numbers preserved.
Failure triage classifies tool errors without auto-retry or permission escalation.
Paired runs show 16 to 25% fewer tool calls and earlier first reads of patched files.
Jev scores never substitute for compilation, tests, or task acceptance.

Watch out

Evidence covers 46 paired tasks at one run each, enough for effort metrics but not outcome claims, and resolve rate is unchanged. 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

typesafe-mod

A Claude Code function hook that sends AskUserQuestion dialogs to Jev and draws a probability per option above the dialog, so you still choose while seeing how confident the model is.

Jev Agents & MCP#community#jevbest#catalog-import

Community

BeLazy167/typesafe-modTypeScript

jev-guard

Claude Code plugin that scores pending tool calls and finished answers with Jev noul probabilities, denying the call with the correction attached when a rule threshold is met.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

muratcakmak/jev-guardTypeScript

langchain-skill-router

LangChain deepagents middleware that replaces SkillsMiddleware with per-turn judging, loading only the SKILL.md files a turn needs so large catalogs stay out of the prompt.

Jev Agents & MCP#community#jevbest#catalog-import

Community

deyna256/langchain-skill-routerPython

jev-mcp

MCP server exposing Jev Choice, Score, Noul, batched ask, file-reading triage, and model listing as typed tools with full probability distributions.

Jev Agents & MCP#community#jevbest#catalog-import

Community

rashedInt32/jev-mcpTypeScript

pi-thinking-router-jev

A pi extension that picks the thinking level for each task, escalating on reasoning-type failures and stepping back down after sustained stability, using Jev choice with local rules as fallback.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

Wh0rigin/pi-thinking-router-jevTypeScript

JevPromptCoach

Claude Code plugin that scores your prompts against seven coding-agent habits on demand rather than on every keystroke, with a backfill so the first report covers prompts you have already written.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

CrowdLinker/JevPromptCoachTypeScript

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai