Can I use riff without paying for Jev calls?

Yes with --no-jev, which skips every semantic rule. Selecting a Jev rule without a key stops with an error instead of silently degrading, so static-only runs stay explicit.

Does riff lock me to Jev, or can I point it elsewhere?

Jev is the default decision model, but any server speaking the System One endpoint works via Pydantic AI decision models. Thresholds are tuned to Jev, so other models need recalibration with the debug output.

Which file types and inputs does it actually read?

Markdown, text, HTML, Word, and PowerPoint, plus fetched article bodies from URLs. Findings carry line and column for text and paragraph or slide labels for Office files.

How do I stop noisy Jev rules from flagging everything I write?

Each Jev finding shows its probability and each rule has its own threshold you can set in config. Run with --debug-jev to see every probability, then raise thresholds or ignore codes, using the lower word floor for decks.

How do I install and run it?

It is not on PyPI yet, so install from source with uv: clone the repo, run uv sync, then uv run riff draft.md. You can also run it without cloning via uvx --from git+https://github.com/scale-venture-partners/riff riff draft.md, or install a persistent command with uv tool install.

What do the exit codes and output formats mean?

Exit code 0 is clean, 1 means findings were reported, and 2 is a usage or file error. Findings print with a rule code, message, and, for Jev rules, the probability; --format json gives machine-readable output, and --list-rules or --explain JEV001 covers the catalog and a single rule.

How does riff decide what text to judge?

It reads each document as a hierarchy of document, section, title, block, and sentence, and every rule declares the level it judges. Most rules read a block, sentence-length reads sentences, and narrative rules sit higher, so one Jev call is made per unit at each level, meaning a section rule costs one call per section rather than per paragraph.

CommunityJev Agents & MCP

Riff

Prose linter that combines millisecond static Python checks with per-judgement Jev probability calls, reporting ruff-style rule codes with tunable thresholds.

See highlights

Open github.com

Stars: 7
Language: Python
License: MIT
Published by: Community
Author: scale-venture-partners
Last verified: Oct 2026

scale-venture-partners/riff

Organization repository on GitHub

View on GitHub

A small, fast prose linter: ruff-style rule codes for writing, backed by TypeSafe's Jev model

Forks: 2
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A small, fast prose linter. It reads a document, flags the writing tells and clarity problems it finds, and reports them with ruff-style rule codes you can select, ignore, and configure. Static rules are pure Python and run in milliseconds; the semantic rules are one call each to TypeSafe's Jev model, which returns a calibrated probability per judgment instead of generated text. riff is not on PyPI yet, so install it from source. It uses uv.

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

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Python

Highlights

Flags writing tells and clarity problems with selectable, ignorable, configurable rule codes.
Asks one yes-or-no Jev question per document unit and prints the probability with each finding.
Reads files as document, section, title, block, and sentence hierarchy with per-level rules.
Installs from source via uv with JSON output, rule catalog, and per-rule explanation commands.

Watch out

Semantic rules require a TypeSafe API key and send document text to the API; without it only static rules run. 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-recruiter

Local LinkedIn recruiting workspace on Browser Use's Jev Ultrafast: Jev screens titles, drives navigation, and assesses profile excerpts against your brief while a live browser stays visible; matches and evidence export as JSON.