What actually works today versus what is still a roadmap goal?

Only the fast read-only check path plus rules list, validate, and Jev-based doctor are exposed. Portable context output, pre-edit guard integrations, pre-push hooks, rule generation, and focused remediation are listed as planned, not as current commands.

Does check run on Node 22, or does it really need Node 26 or newer?

Node 26 or newer, plus TYPESAFE_API_KEY and a Git repository. Install with npm install --global neuralint@alpha, or from a checkout with pnpm install, pnpm check, pnpm build, and npm link, and pin neuralint@0.1.0-alpha.3 in automation while the package is alpha. TYPESAFE_API_URL can select a compatible endpoint; Pi credentials for GPT-6 Astra are only needed for rules doctor --explain.

How should I read the rules doctor advisory and failure bands?

An advisory sits at 45% failure probability and a failure at 75%, as fixed product thresholds. The README notes a 16-rule calibration corpus left the failure band provisional, so treat doctor output as guidance and keep thresholds in your rules.

Will check flood CI with inconclusive findings or silently drop cases?

By default check prints only definitive matches; add --advisories for inconclusive ones and --format json for machines. Limits such as maxFiles, maxCases, and maxInputTokens fail before inference, and runs never silently truncate cases.

What Node version does neuralint need?

Node.js 26 or newer, plus a TypeSafe API key and a Git repository; rules doctor --explain additionally needs Pi credentials with access to GPT-6 Astra for the optional explanation.

How do I bootstrap neuralint in a repository?

neuralint init --base origin/main creates .neuralint/config.yaml and an example rule under .neuralint/rules/, idempotently and without overwriting existing configuration, after which you edit the rule and run rules validate, rules list and check.

What resource budgets does neuralint enforce?

Config limits such as maxFiles 500, maxCollectionBytes 5,000,000, maxCases 1000, maxRequests 20 and maxInputTokens 500,000 fail before inference when exceeded; neuralint never silently truncates cases to fit a run budget.

CommunityJev Tools & CLIs

Neuralint

Diff-scoped semantic linter that keeps versioned YAML rules in-repo, checks merge-base diffs with bounded Jev decision matrices, and emits findings a coding agent can act on.

See highlights

Open github.com

Stars: 1
Language: HTML
License: MIT
Published by: Community
Author: jpowersdev
Last verified: Oct 2026

jpowersdev/neuralint

User repository on GitHub

View on GitHub

AI code review against your repository's rules, powered by TypeSafe AI's Jev.

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Fast semantic linting for changed code, powered by TypeSafe AI's Jev. neuralint lets a repository keep versioned engineering rules alongside its code. It compares the selected base ref with HEAD, evaluates applicable rules against changed code in a bounded Jev decision matrix, and returns localized candidate violations that a coding agent can act on immediately. The intended product carries repository-owned guidance through the lifecycle of a change:

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

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Highlights

Repository-owned rules under .neuralint/rules with strict schemas, duplicate-ID checks, and example validation.
Merge-base-aware diffs with deterministic routing, changed-span locations, and optional Tree-sitter scopes.
Jev packs bounded by 64k aggregate and 32k binding budgets with concurrent requests and oversize splitting.
check supports --rule, --stdin, --format json, failOn severities, and preflight cost and size limits.

Watch out

Alpha: CLI, rule and JSON report formats may change before 1.0, and context, guard, and remediation commands are still goals. 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

SwiftJev

Swift provider adapter that sends typed noul, choice, and score prompts to the hosted System One API through SwiftDecision DecisionBackend, returning validated probability distributions.