Why use a model instead of plain edit-distance correction?

Edit distance only sees spelling, while this sends candidate names plus help descriptions to Jev as a Choice question. That is why meaning-close but spelling-far pairs such as remove to rm can resolve.

Does every command I type incur a network round trip?

Valid commands never touch the network; only the unknown-token path sends one Choice request. The hot path is a single cached JSON read, and offline edit-distance matching covers key or API failures.

How do I stop it from auto-running something destructive?

Prompt mode lists up to three candidates above threshold and waits for one keypress, hint mode runs as typed, and auto mode only runs without asking above a high threshold with destructive names denylisted.

What is the safe way to install it without slowing my shell?

Add one eval line per shell for named commands, or use the mise alias snippet. Avoid the all-binaries mode: it misreads filenames as subcommands, adds startup cost, and prompts far too often.

How do I install it and wrap a command?

Run go install github.com/syumai/jevyoumean/cmd/jym@latest, then wrap commands with the optional -- separator, as in jym -- git switch main. For a project-local alias, put the [shell_alias] snippet from jym --print-mise into mise.toml, or generate shell functions with jym --shell-integration bash or zsh.

What exactly does it send to Jev?

The command name, the subcommand path, the mistyped token, and the candidate subcommand names plus their help descriptions. Arguments are not sent by default because context_args is none; flags sends flag names only and all sends everything. Wrapping an internal CLI sends its subcommand structure to TypeSafe.

What happens when the API is unavailable?

Without a key or on API failure, jym falls back to edit-distance matching (distance at most 2 or len/3, plus prefix matches), shown without probabilities and marked (offline). When Jev answers, even with its __none__ escape hatch, that verdict stands and no fallback runs.

CommunityJev Agents & MCP

Jevyoumean

Shell wrapper that intercepts unknown subcommands, reads the target CLI help output, and asks Jev which documented subcommand matches the intent behind the typo.

See highlights

Open github.com

Stars: 16
Language: Go
License: MIT
Published by: Community
Author: syumai
Last verified: Oct 2026

syumai/jevyoumean

User repository on GitHub

View on GitHub

Semantic "Did you mean?" for any CLI — wraps commands and uses TypeSafe's Jev to match subcommand typos by intent, not edit distance.

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

jym wraps any CLI command. When you enter a subcommand the CLI does not document, it reads the CLI's help output and asks Jev — TypeSafe's System One model — which documented subcommand you most likely meant. Unlike an edit-distance Did you mean?, jym matches on intent: it hands Jev the candidate subcommand names plus their help descriptions. remove → rm, list → ps, undo → restore are close in meaning but far in spelling — that is the gap this experiment targets.

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

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification Jev for Routing

Used inJev for Software & DevTools

Languages: Go

Highlights

Matches on intent from help text, not spelling, with a mandatory none-of-these escape option in each request.
Check-first design inspects arguments before executing, so side effects can never run twice.
Prompt, hint, and auto modes with thresholds, denylist, TTY fallback, and offline edit-distance matching.
Help-output cache keyed by binary path and metadata with weekly expiry, plus per-project config and learned tokens.

Watch out

The author calls it experimental and unsuitable for production-critical use, and wrapping every binary on PATH is explicitly discouraged. 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

jev-mcp

Go MCP server exposing Jev through OpenRouter or TypeSafe as five tools: decide, classify, verify, rerank and list_models. Each tool call is one paid inference request, answers are validated against the question asked, and the unchanged provider response is returned with its usage cost.