Multi-Agent Modes

Overview of PuzldAI’s collaborative AI workflows

PuzldAI supports multiple modes for orchestrating AI models to work together on tasks.

Available Modes

Compare

Run the same prompt on multiple models and compare responses side-by-side.

Debate

Multi-round discussion where models argue and refine their positions.

Correction

One model produces, another reviews and optionally fixes.

Consensus

Models propose solutions, vote on the best approach, and synthesize a final answer.

When to Use Each Mode

Mode	Best For
Compare	Getting diverse perspectives, choosing between approaches
Debate	Exploring trade-offs, uncovering edge cases
Correction	Quality assurance, code review, fact-checking
Consensus	Complex decisions requiring agreement, architectural choices

Quick Examples

Compare Mode

/compare claude,gemini "Write a function to parse JSON safely"

Debate Mode

/debate claude,gemini "Should we use GraphQL or REST for this API?"

Correction Mode

/correct claude gemini "Implement binary search"

Consensus Mode

/consensus claude,gemini,codex "Design the database schema for a todo app"

Background Execution

All multi-agent modes support background execution:

• Press Escape to hide the view and return to chat (execution continues)
• Press Ctrl+E to return and check progress
• Press Ctrl+C to cancel the operation

A status indicator shows progress when running in the background.