multiarena

English | 中文

Stop relying on a single AI's opinion. multiarena lets you talk to multiple LLMs at once — Claude, GPT, Gemini, DeepSeek, and more — compare their answers, and let them debate each other to produce the best result.

The Problem

Every AI model has blind spots. Ask one model to write a brand story, and you get a decent draft. Ask one model to analyze a decision, and you get one perspective. The best work comes from multiple viewpoints — but switching between chat tabs is painful.

The Solution

multiarena puts multiple models in the same room. Broadcast a question, see every model's answer side-by-side. Switch to team mode, and models collaborate in relay — drafting, revising, polishing — to produce a document that's been through 4 pairs of eyes, not just one.

What You Can Do

| Task | How multiarena helps | |------|---------------------| | Write better copy | 3 models draft a brand story → debate each other's choices → deliver a polished piece | | Analyze decisions | Ask all models to evaluate a proposal → compare their reasoning → spot consensus and blind spots | | Review documents | Models take turns critiquing your draft from different angles (skeptic, pragmatist, user advocate) | | Brainstorm ideas | Broadcast a prompt → 4 models each contribute unique angles → merge into one document |

Web UI

Clean, intuitive interface for non-technical users. Zero config — start with a setup wizard.

┌──────────────────────────────────────────────────────────┐
│  ⚡ multiarena                           [⚙] [🌙] [EN]  │
│──────────────────────────────────────────────────────────│
│                                                          │
│               Good afternoon, Tim 👋                      │
│          What would you like to do today?                 │
│                                                          │
│  ┌── ✍️ Write ────┐ ┌── 📊 Analyze ────┐               │
│  │ Brand story    │ │ Compare options  │               │
│  │ Ad copy        │ │ Analysis report  │               │
│  │ Product intro  │ │ Data insights    │               │
│  └───────────────┘ └──────────────────┘               │
│                                                          │
│  ┌── 💡 Ideate ───┐ ┌── 🔍 Review ─────┐               │
│  │ Product name   │ │ Find gaps         │               │
│  │ Slogan         │ │ Improve wording   │               │
│  │ Campaign ideas │ │ Polish            │               │
│  └───────────────┘ └──────────────────┘               │
│                                                          │
│  ┌──────────────────────────────────────────────────┐   │
│  │  Write a brand story for my specialty coffee     │   │
│  │  roastery, for our website About page…           │   │
│  └──────────────────────────────────────────────────┘   │
│                              [Start AI Team →]           │
└──────────────────────────────────────────────────────────┘

→ multiarena (no args needed)

CLI

Terminal-native for developers. Keyboard-driven, fast, stays in your workflow.

┌─ claude ──────────────────┐ ┌─ gpt ─────────────────────┐
│                           │ │                           │
│  Let me analyze this      │ │  I'd evaluate this from   │
│  approach in detail:      │ │  three angles:            │
│                           │ │                           │
│  Strengths:               │ │  1. Technical feasibility │
│  1. Clean four-layer arch │ │  2. Maintenance cost      │
│  2. Simple Provider API   │ │  3. Room for expansion    │
│                           │ │                           │
│  2K/200K · 15 lines · done│ │  1K/128K · 12 lines · done│
└───────────────────────────┘ └───────────────────────────┘
──────────────────────────────────────────────────────────
 claude ●  gpt ●  — Tab:model d:compare m:mute q:quit
[all] > ▊

→ multiarena terminal

Team deliberation: models challenge each other. Draft → Revise → Polish → Review — 4 perspectives fight over every sentence. The result survives real scrutiny.

Quick Start

npm install -g multiarena
multiarena                # Opens Web UI in your browser
multiarena terminal       # Terminal mode (for CLI lovers)

No config? No problem. The Web UI has a setup wizard. Or create .multiarenarc:

[models.claude]
provider = "anthropic"
model = "claude-sonnet-4-6"
api_key = "${ANTHROPIC_API_KEY}"

[models.gpt]
provider = "openai"
model = "gpt-4o"
api_key = "${OPENAI_API_KEY}"

[defaults]
active = ["claude", "gpt"]

Installation

npm install -g multiarena          # Recommended
git clone [email protected]:timgunnar/multiarena.git  # From source

Supported Models

Anthropic (Claude) · OpenAI (GPT-4o) · Google (Gemini) · DeepSeek · MiniMax · Ollama (local)

CLI

multiarena              # Web UI (no args needed)
multiarena terminal     # Terminal mode
multiarena -r <id>      # Resume session
multiarena -l           # List sessions
multiarena -h           # Help

License

MIT

📖 User Guide — Full documentation: architecture, configuration, Web UI, session management, troubleshooting