How does it compare to Ruff and to LLM PR reviewers?

Deterministic linters are instant but syntax-bound, LLM reviewers are deep but slow and costly, while klint does semantic checks in one forward pass with zero decoded tokens.

Can I run audits fully offline without the Jev API?

Yes. Default local mode runs open-weight kev models in-process, while remote mode points at a Jev endpoint or self-hosted kev.serve for hosted inference.

How do I write custom architectural rules?

Phrase instructions as a positive practice rather than an antipattern, set target unit types, and merge or replace the built-in backend, React and project packs.

How do I keep noisy findings out of CI results?

It emits Pass, Fail, Irrelevant or Lack of Evidence sorted by confidence, with min-confidence filtering and JSON output for CI pipelines.

What do I need installed for the default local mode?

Python 3.11 or newer plus kev, installed with pip install "kev[serve] @ git+https://github.com/jaredpalmer/kev.git". The zero-config default runs local in-process jaredpalmer/kev-4b, and the open-weight kev-0.8b and kev-4b models are the offline path. Remote mode points at any /v1/systemone endpoint instead.

What does klint-project add beyond file-level checks?

It audits Python package and directory hierarchies layer by layer, combining AST-based file metadata such as imports, exports, classes, functions, line_count and __init__.py public API status with recursive subpackage dependency rollups and mutual cycle detection.

Why must a klint rule be phrased as a positive practice?

Because klint judges Pass when code follows a practice and Fail when it violates it, the config tells you to write each instruction as " rather than ?", for example "Does the code use a structured logger rather than raw print() statements for application logging?"

CommunityJev Models & Reimplementations

Klint

Python CLI that audits files and package trees against plain-English JSON rule packs in one System One forward pass, locally with kev or remotely with Jev.

See highlights

Open github.com

Stars: 5
Language: Python
License: APACHE-2.0
Published by: Community
Author: skydiving94
Last verified: Oct 2026

skydiving94/klint

User repository on GitHub

View on GitHub

Declarative semantic linter and architectural code auditor for local open-weight kev and remote System One (/v1/systemone) models.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

klint bridges the gap between rigid, syntax-bound linters (like ESLint or Ruff) and slow, expensive LLM code reviewers (like PR bots streaming tokens for 40 seconds). By leveraging the System One (/v1/systemone) pointer-head specification, klint evaluates files and recursive package trees against custom JSON rule packs in a single forward pass with zero decoded tokens.

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

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Python

Highlights

Evaluates a whole rule pack in one forward pass with zero decoded tokens.
Runs 100% local on kev or remote on Jev and self-hosted kev.serve endpoints.
Audits single files, directory trees and recursive Python package structure.
Classifies Pass, Fail, Irrelevant or Lack of Evidence sorted by confidence.

Watch out

Semantic judgments depend on the configured kev or Jev host; phrase instructions as positive practices and filter by confidence. 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

gut

Python helper that turns judgment calls into one-liners like likely, classify and rate, defaulting to Jev while also running on local or OpenAI-compatible backends.