grupr — multi-LLM code review CLI

Code review is a panel sport, not a solo sport.

npx @grupr/cli review my-feature.diff

5 expert Skills (Architect · Security · Performance · Maintainability · Synthesizer) review your code in parallel. One unified verdict. One verified patch.

Install

# One-off
npx @grupr/cli review my-feature.diff

# Or globally
npm install -g @grupr/cli
grupr review my-feature.diff

Requires Node 18+.

Quick start

# 1. Get an API key (first time only)
grupr login

# 2. Review a diff with the default 5-Skill panel, Deep tier
grupr review my-feature.diff

# Or pipe from git
git diff main..feature | grupr review --output verified.patch

The default panel runs all 5 launch Skills, each backed by the model that's strongest at that lens (Anthropic, OpenAI, Groq). The Synthesizer reads everyone's verdict and produces one unified output. Deep tier (the default) also generates a sandboxed, test-verified patch.

Free tier: 2 Quick reviews + 1 Deep review on us, no card. Want more? Add your own API keys in Settings → API Keys or upgrade to Pro.

Usage

grupr review [file] [options]

OPTIONS
  --tier <quick|deep>     Review tier. Default: deep.
  --output, -o <file>     Write the verified patch diff to <file>.
  --skills <list>         Comma-separated Skill IDs to run. Default: all 5.
  --api-base <url>        API base URL. Default: https://api.grupr.ai.
  --api-key <key>         Bearer token. Defaults to GRUPR_API_KEY env var.
  --quiet, -q             Suppress progress output.

Examples

# Just the Synthesizer's verdict on a PR diff
git diff origin/main | grupr review --tier quick

# Full Deep-tier review, save the patch
grupr review feature.diff --output verified-patch.diff

# Only run two Skills (security + performance):
grupr review feature.diff --skills security,performance,synthesizer

# Run against a self-hosted Grupr instance
grupr review feature.diff --api-base https://grupr.your-company.com

Skills

A Skill is a small JSON manifest combining a system prompt, a recommended model, and an output contract. The default panel is the 5 launch Skills:

Skill	Model
Architect	Claude Opus 4
Security	GPT-4o
Performance	Llama 3.3 70B (Groq)
Maintainability	Llama 3.3 70B (Groq)
Synthesizer	Claude Opus 4

Pick a subset with --skills architect,security. Bring your own with the open registry at grupr-ai/skills — PR your manifest, get included by default.

Library use

The CLI's API surface is also exported for programmatic use:

import { Client } from 'grupr';

const c = new Client({ accessToken: process.env.GRUPR_API_KEY! });
const created = await c.createCodeReview({
  tier: 'deep',
  code: diffText,
});
// poll, approve, fetch patch …

For agent-side use (build an agent that participates IN Grupr conversations), use @grupr/sdk instead. This package (@grupr/cli) is for the user-side review experience.

License

MIT — see LICENSE.