consensus

A CLI tool that sends prompts to multiple AI providers (OpenAI, Anthropic, Gemini) concurrently and saves their responses for comparison. Includes optional email notifications for real-time updates and a web UI for interactive use.

Requirements

• Go 1.24 or higher
• Environment variables:
  • OPENAI_API_KEY for OpenAI API
  • ANTHROPIC_API_KEY for Anthropic API
  • GEMINI_API_KEY for Google Gemini API

Configuration

The app uses a YAML configuration file for email settings. On first run, a default config file is created at:

• $XDG_CONFIG_HOME/consensus/config.yml (if XDG_CONFIG_HOME is set)
• ~/.config/consensus/config.yml (on most systems)
• config.yml (fallback in current directory)

Default Configuration

email:
  smtp_host: smtp.gmail.com
  smtp_port: 587
  from_email: consensus.ai.25@gmail.com
  from_name: Consensus AI
  password_env_var: CONSENSUS_EMAIL_PASSWORD
  subject_prefix: "[Consensus AI]"

providers:
  - name: openai
    type: openai
    api_key_variable: OPENAI_API_KEY
    model: gpt-4o
  - name: gemini
    type: gemini
    api_key_variable: GEMINI_API_KEY
    model: gemini-2.0-flash
  - name: anthropic
    type: anthropic
    api_key_variable: ANTHROPIC_API_KEY
    model: claude-4-sonnet-20250514
    max_tokens: 64000

prompt_provider: openai
response_providers:
  - openai
  - gemini
  - anthropic

Configuration Options

Providers: Configure which AI providers are available and their settings

• name: Unique identifier for the provider
• type: Provider type (openai, gemini, anthropic)
• api_key_variable: Environment variable containing the API key
• model: Model to use for this provider
• max_tokens: Optional token limit (primarily for Anthropic)
• base_url: Optional custom API endpoint

Provider Selection:

• prompt_provider: Which provider to use for optimizing prompts (default: openai)
• response_providers: List of providers to generate responses (default: all three)

Optional Email Notifications

• Environment variable for email password (configurable via password_env_var in config, defaults to CONSENSUS_EMAIL_PASSWORD)

Usage

Interactive Mode

go run main.go

The tool will prompt you to enter your request, then process it through all AI providers.

Command Line Mode

# Using full flag name
go run main.go -prompt "Compare the pros and cons of React vs Vue"

# Using shorthand
go run main.go -p "What are the latest trends in AI?"

# With email notifications
go run main.go -prompt "Your prompt here" --email-to "user1@example.com,user2@example.com"

# Email shorthand
go run main.go -p "Your prompt here" -e "user1@example.com,user2@example.com"

# Disable master prompt optimization
go run main.go -p "Your prompt here" --no-master-prompt
go run main.go -p "Your prompt here" -nmp

# Override which provider handles master prompt optimization
go run main.go -p "Your prompt here" --master-prompt-provider "anthropic"
go run main.go -p "Your prompt here" -mpp "gemini"

# Override which providers generate responses
go run main.go -p "Your prompt here" --response-providers "openai,gemini"
go run main.go -p "Your prompt here" -rp "anthropic,openai"

# Combined configuration overrides
go run main.go -p "Your prompt here" -mpp "gemini" -rp "openai,anthropic" -e "user@example.com"

Server Mode (Web UI)

# Start the server with web UI (default port 8080)
go run main.go --serve

# Use a custom port
go run main.go --serve --port 3000

# Keep server running for multiple requests (default shuts down after first request completes)
go run main.go --serve --multi-session

The server mode automatically opens your default browser to the web interface. The UI allows you to:

• Enter prompts interactively
• Select which providers to use for responses
• Toggle master prompt optimization
• View responses from all providers side by side

How it works

The consensus tool follows these steps:

1. Accepts user input via command line flags (-prompt or -p) or interactive stdin prompt
2. Uses configurable provider (default: OpenAI) with a master prompt (C.R.A.F.T. methodology) to optimize the user's request (can be disabled with --no-master-prompt or overridden with --master-prompt-provider)
3. Sends the optimized prompt concurrently to configured AI providers (configurable via --response-providers or config file)
4. Saves all outputs to the responses/ directory with UUID-based filenames:
   • id-{uuid}-request.txt - Original user request
   • id-{uuid}-prompt.txt - Optimized prompt created by master prompt provider
   • id-{uuid}-{Provider}.txt - Each provider's response (e.g., OpenAI.txt, Anthropic.txt, Gemini.txt)
5. Optionally sends HTML email notifications for each step when email is configured

Architecture

• Provider Interface: All AI providers implement a consistent Provider interface
• Output Manager: Flexible output system supporting multiple writers (file and email)
• UUID Sessions: Each run generates a unique session ID for organized file storage and email tracking
• Concurrent Processing: All AI providers are queried simultaneously for faster results
• Email Notifications: Optional real-time email updates with HTML formatting and provider-specific styling
• Embedded Web UI: Self-contained HTTP server with an embedded web interface for interactive use

Development

Using Make (Recommended)

# Build and run
make run

# Run with a specific prompt
make run PROMPT="Your prompt here"

# Run tests
make test

# Build binary
make build

# Clean build artifacts
make clean

Direct Go Commands

# Build and run
go run main.go

# Build and run with prompt
go run main.go -prompt "Your prompt here"

# Start web UI server
go run main.go --serve

# Run tests
go test ./...

# Build binary
go build