claude-consensus

Multi-model code review and plan review for Claude Code

Multiple AI models independently review code or plan implementations, then converge on consensus through structured synthesis and approval rounds. Configurable — works with as few as Claude + 1 external model.

Quick Start

Just tell Claude:

"Set up claude-consensus: https://github.com/AltimateAI/claude-consensus"

That's it for most users. The rest of this README is for understanding what's happening under the hood.

Manual Setup

# 1. Install the plugin
/plugin marketplace add AltimateAI/claude-consensus
/plugin install consensus

# 2. Configure your models
/consensus-setup

# 3. Use it
/code-review "staged changes"
/plan-review "Add caching to the API layer"

Prerequisites

Requirement	Required?	Notes
Claude Code CLI	Yes	The host environment
Kilo CLI	Recommended	Routes all 7 models via OpenRouter with 1 API key
OpenRouter API key	Recommended	Required if using Kilo CLI
Native CLIs (codex, gemini)	Optional	Alternative to OpenRouter for specific models

Minimal setup: Claude + 1 external model is enough for consensus reviews.

Supported Models

Model	Provider	OpenRouter ID	Native CLI
Claude	Anthropic	(built-in)	Claude Code
GPT 5.2 Codex	OpenAI	openai/gpt-5.2-codex	codex
Gemini 3.1 Pro	Google	google/gemini-3.1-pro-preview	gemini
Kimi K2.5	Moonshot	moonshotai/kimi-k2.5	—
Grok 4	xAI	x-ai/grok-4	—
MiniMax M2.5	MiniMax	minimax/minimax-m2.5	—
GLM-5	Zhipu AI	z-ai/glm-5	—
Qwen 3.5 Plus	Alibaba	qwen/qwen3.5-plus-02-15	—

Installation

From GitHub

# Via CLI
claude plugin marketplace add AltimateAI/claude-consensus
claude plugin install consensus

# Or in-session
/plugin marketplace add AltimateAI/claude-consensus
/plugin install consensus

From Source (local clone)

git clone https://github.com/AltimateAI/claude-consensus.git

Then in a Claude Code session:

/plugin marketplace add /path/to/claude-consensus
/plugin install consensus

Configuration

Setup Wizard (Recommended)

Run /consensus-setup to interactively configure:

• Provider: OpenRouter (1 key for all models), native CLIs, or both
• API key: Saved to ~/.claude/.env
• Models: Choose which of the 7 external models to enable
• Quorum: Minimum models that must respond for a valid review
• Smoke test: Verifies each model responds before finalizing

Config Location

Path	Purpose
~/.claude/consensus.json	User config (created by /consensus-setup)
plugins/consensus/consensus.config.json	Plugin defaults (fallback)

Config Format

{
  "version": "1.0.0",
  "models": [
    {
      "id": "gpt",
      "name": "GPT 5.2 Codex",
      "command": "kilo run -m openrouter/openai/gpt-5.2-codex --auto",
      "resume_flag": "-c",
      "enabled": true
    }
  ],
  "min_quorum": 5
}

Manual Configuration

Edit ~/.claude/consensus.json directly to:

• Enable/disable models ("enabled": true/false)
• Change CLI commands
• Adjust quorum

Or re-run /consensus-setup anytime to reconfigure.

Commands

Command	Purpose
/consensus-setup	Configure models, API keys, and quorum
/code-review [target]	Multi-model code review with consensus
/plan-review [task]	Multi-model implementation planning with consensus

Usage

# Code review
/code-review "staged changes"
/code-review PR #123
/code-review "last 3 commits"
/code-review app/service/foo.py

# Plan review
/plan-review "Add caching to the API layer"
/plan-review "Refactor auth to use JWT"

How It Works

┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐
│ Claude  │  │ Model 1 │  │ Model 2 │  │ Model N │
└────┬────┘  └────┬────┘  └────┬────┘  └────┬────┘
     │            │            │            │
     ▼            ▼            ▼            ▼
  Phase 1: Independent Review/Planning (parallel)
     └────────────┴────────────┴────────────┘
                       │
                       ▼
            Phase 2: Synthesis
         (consensus, conflicts, comparison table)
                       │
                       ▼
            Phase 3: Convergence
         (APPROVE / CHANGES NEEDED, max 2 rounds)
                       │
                       ▼
              Final Result with Attribution

• Quorum: Configurable (default: 5). Strict majority of participants must respond.
• Graceful degradation: Unavailable models are skipped at runtime if quorum is still met.
• Session artifacts: Saved in /tmp/ for debugging.
• Independent reviews: Each model reviews with no cross-contamination.

Contributing

Contributions are welcome! Please read the Contributing Guide before submitting a PR.

This project follows the Contributor Covenant Code of Conduct.

License

MIT License. Copyright (c) 2026 Altimate AI.