User repository on GitHub
Jev-powered tool and skill selection, context search, and output triage for Codex via MCP
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Agents & MCP
Jev in Codex
Codex plugin and MCP tool that batch-labels JSONL text records with Jev: one jev_label tool assigns a label per record and hands low-confidence decisions back to Codex for review. Its benchmark reports less Codex input and elapsed time on synthetic workloads.
- Stars
- 6
- Language
- TypeScript
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- teempai
- Last verified
- Oct 2026
- Languages
- TypeScript
- Platforms
- MCP
- Capabilities
- Open sourceRequires API key
Highlights
- One tool, jev_label: it turns a JSONL file of text records into a labelled file, one label per record, with custom policies plus a feedback_theme preset.
- On 64-record synthetic workloads the 0.3 benchmark reports 21.5–31.9% less Codex input and 8.4–23.5% less elapsed time over four paired repetitions.
- The earlier 0.2 benchmark on 64 synthetic feedback messages measured 42.7% less Codex input and 26.4% less elapsed time; tiny batches did worse.
- Decisions below 0.8 confidence come back with the original text so Codex can review and correct the file; existing output files are never overwritten.
- Input limits: 1–256 records, IDs up to 100 letters/digits/underscore/hyphen, nonempty text up to 6,000 characters, no extra fields, 1 MiB files.
Quickstart
json
{"path":"feedback/items.jsonl","policy":"feedback_theme"}Watch out
Copy reviewed against the repo README on 2026-10-15. The GitHub description still advertises tool and skill selection, context search and output triage, but the README says those tools were removed because their tested workflows showed no useful improvement. Benchmarks are exploratory verification on 64-record synthetic development data, not independent confirmation.
Questions about this entry
Short answers drawn from the linked project or page.
It is a Codex plugin for batch labelling with Jev, delivered as an MCP tool. Feed it an approved UTF-8 JSONL file of text records and one of its policies, and it writes a decisions.jsonl with every original ID and exactly one label per record, while Codex reviews the uncertain ones.
The project's benchmark claims less work for Codex: on 64-record synthetic workloads the 0.3 custom-policy benchmark measured 21.5–31.9% less total Codex input and 8.4–23.5% less elapsed time across four paired repetitions, with both arms labelling all records correctly. It is exploratory verification, not independent confirmation.
Four: reliability, usability, pricing and feature. The shipped example labels a report-disappearing complaint as reliability and a request for a cheaper plan as pricing. Custom question and criteria policies let you define your own labels for topics, sentiment or document collections.
Ask Codex to install the repository using docs/INSTALL.md, configure your workspace and TypeSafe key privately, and verify that only jev_label is available. Node.js 22+ is required; build with npm ci --ignore-scripts && npm run check, set TYPESAFE_API_KEY privately, and pass --root or JEV_WORKSPACE_ROOT.
Input accepts 1–256 records per file with unique IDs of up to 100 letters, digits, underscore or hyphen, nonempty text up to 6,000 characters per record and no extra fields; files are limited to 1 MiB. Both input and output stay inside the configured workspace, existing files are never overwritten, and the output parent must exist.
No. The README calls them exploratory verification on 64-record synthetic development workloads and states they do not prove real-world accuracy, general labelling superiority, invoice savings or automatic desktop adoption. Tiny batches performed worse in an earlier sweep, and user-supplied taxonomies still need validation on actual data.
The README states the model is pinned to the benchmarked jev-1.13.0 and also flags the old JEV_MODEL override; check docs/INSTALL.md for the current behaviour, because the captured text cuts that sentence short. New capabilities must pass a reviewable benchmark against plain Codex before being exposed.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Jev in front of the HeyGen MCP
X: Playwright CLI + Jev vs Playwright MCP
X: A local MCP server that cuts computer use from 23s to 0.8s. Cartwright built Cortex, a local MCP server where…
X: A side panel that drives any site's WebMCP tools
More Jev resources like this
A self-hosted MCP gateway on Cloudflare Workers that wraps the TypeSafe/Jev System One model into nine judgment tools for intent routing, retrieval reranking and batch semantic work, with a D1-backed admin console.
Jev Agents & MCP#community#catalog-import#logicrw
Open-source Codex plugin that consults TypeSafe Jev about tools, models, tasks, skills, context, and strategies, diagnoses failed commands, and checks completion claims against evidence. The current package is version 0.4.0.
Jev Agents & MCP#community#catalog-import#logicrw
Six evidence-oriented tools for the pi and omp coding agents, also usable from any MCP client: navigate unfamiliar code, ask typed questions about repository evidence, review completed changes, and select existing tests.
Jev Agents & MCP#community#catalog-import#logicrw
Community
nomenak/jev-toolsTypeScript
nomenak/jev-toolsTypeScriptMCP server that lets Claude Code, Codex and other agents drive a real Chrome browser while JEV makes the small typed decisions — which element, which suggestion, did the click work — with calibrated confidence driving act, verify or ask.
Jev Agents & MCP#community#catalog-import#logicrw
Community
legostin/jev-mcpTypeScript
legostin/jev-mcpTypeScriptAn MCP server that gives coding agents only the JS/TS code a task needs: the TypeScript compiler follows calls, types, callers and tests, Jev judges relevance, and results come back within a token budget.
Jev Agents & MCP#community#catalog-import#logicrw
Community
123wwwa/jevtraceTypeScript
123wwwa/jevtraceTypeScriptA dependency-free MCP server that exposes TypeSafe's Jev to Claude Code, Codex, Cursor, Hermes, and other agents through the ask_jev and list_jev_models tools, so your code gets typed decisions with probabilities instead of prose to parse.
Jev Agents & MCP#community#catalog-import#logicrw
Community
codeia-academy/jev-mcpPython
codeia-academy/jev-mcpPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.