verdict

Multi-agent AI debate engine for your terminal. Three AI agents argue both sides, then deliver a verdict.

Zero signup. BYOK. Instant decisions.

npx askverdict "Should I use PostgreSQL or MongoDB for my SaaS?"

  THE ADVOCATE (Claude claude-sonnet-4-5)                           FOR
  PostgreSQL's ACID compliance and mature ORM ecosystem make it
  the default choice for any SaaS handling financial data...

  THE CRITIC (GPT-4o)                                       AGAINST
  MongoDB's flexible schema lets you ship MVPs in days, not
  weeks. Schema migrations are the #1 velocity killer...

  +-----------------------------------------------------+
  |  VERDICT                                        78%  |
  |  PostgreSQL is the stronger choice for most SaaS     |
  |  applications, especially those handling payments    |
  |  or relational data.                                 |
  |                                                      |
  |  For:  ACID compliance, mature ORM ecosystem         |
  |  Against: MongoDB's flexible schema for MVPs         |
  |  Blind spots: Multi-region replication costs         |
  |  Next: Benchmark your actual query patterns          |
  +-----------------------------------------------------+

  3 providers | 2,847 tokens | 12.3s

Quick start

You need at least one API key:

# Set one or more (env vars)
export ANTHROPIC_API_KEY=sk-ant-...
export OPENAI_API_KEY=sk-...
export GEMINI_API_KEY=AI...

# Run it
npx askverdict "Should I use Kubernetes or just Docker Compose?"

If you have multiple keys, each debate agent uses a different provider for diverse perspectives.

Install (optional)

npm i -g askverdict

Usage

# Basic debate
verdict "Microservices vs monolith for a 5-person startup?"

# Fast mode (2-3 points, ~10s)
verdict "Rust or Go for CLI tools?" --mode fast

# Thorough mode (5-7 points, ~30s)
verdict "Should we raise a Series A now?" --mode thorough

# Force specific providers
verdict "React vs Vue?" --providers anthropic,openai

# Machine-readable JSON output
verdict "Buy vs rent in Austin?" --json

# Disable streaming
verdict "Hire or outsource?" --no-stream

Configuration

# See detected API keys
verdict config

# Save keys to ~/.verdict/config.json (no env vars needed)
verdict config --set anthropic=sk-ant-...
verdict config --set openai=sk-...
verdict config --set google=AI...

# Show config file path
verdict config --path

Environment variables always take priority over the config file.

How it works

1. The Advocate argues FOR your proposition with evidence and data
2. The Critic argues AGAINST with risks, counterexamples, and failed implementations
3. The Synthesizer reads both sides, picks a winner, and delivers a verdict with confidence score

Advocate and Critic run in parallel, so debates complete in roughly the time of one LLM call.

Modes

| Mode | Points | Tokens | Time | |------|--------|--------|------| | fast | 2-3 | ~1,500 | ~10s | | balanced | 3-5 | ~3,000 | ~15s | | thorough | 5-7 | ~6,000 | ~30s |

Provider priority

If you have multiple API keys, verdict distributes agents across providers:

| Keys available | Advocate | Critic | Synthesizer | |---------------|----------|--------|-------------| | 3 keys | Provider 1 | Provider 2 | Provider 3 | | 2 keys | Provider 1 | Provider 2 | Provider 1 | | 1 key | Same | Same | Same |

JSON output

verdict "PostgreSQL vs MySQL?" --json

{
  "question": "PostgreSQL vs MySQL?",
  "mode": "balanced",
  "advocate": {
    "provider": "anthropic",
    "model": "claude-sonnet-4-5-20250514",
    "argument": "...",
    "tokens": 1200,
    "durationMs": 4500
  },
  "critic": {
    "provider": "openai",
    "model": "gpt-4o",
    "argument": "...",
    "tokens": 1100,
    "durationMs": 3800
  },
  "verdict": {
    "summary": "PostgreSQL is the stronger choice...",
    "confidence": 78,
    "forPoints": ["..."],
    "againstPoints": ["..."],
    "blindSpots": ["..."],
    "nextStep": "Benchmark your query patterns"
  },
  "meta": {
    "totalTokens": 3500,
    "durationMs": 8200,
    "providers": ["anthropic", "openai", "google"],
    "version": "0.1.0"
  }
}

Want more?

AskVerdict AI is the full platform with:

• 5+ specialized personas (Engineer, Economist, Analyst, and more)
• Multi-round debates with follow-up questions
• Saved history and shareable verdict URLs
• Team collaboration and decision tracking
• Domain-aware persona selection
• No API keys needed (managed credits or BYOK)

License

MIT