Module patterns 

Multi-agent pattern primitives

Built-in multi-agent collaboration patterns that extend beyond simple swarm/supervisor modes. Each pattern implements a distinct collaboration strategy:

• Debate — Multiple agents argue different positions, then converge
• Review-Loop — One agent produces, another reviews, iterating to quality
• Research-Synthesize — Parallel research agents feed a synthesizer
• Voting — Multiple agents vote on options, majority decides

These patterns are first-class modes accessible via --mode debate, --mode review-loop, --mode research-synthesize, or --mode voting.

Structs

PatternConfig

Configuration for multi-agent patterns

Functions

run_debate

Run a debate between multiple agents with different positions.

run_debate_multi

Run debate mode with multiple LLM providers (round-robin)

run_research_synthesize

Run parallel research agents followed by a synthesizer.

run_research_synthesize_multi

Run research-synthesize mode with multiple LLM providers

run_review_loop

Run a producer-reviewer loop that iterates until quality is met.

run_review_loop_multi

Run review-loop mode with multiple LLM providers

run_voting

Run a voting process where multiple agents evaluate options.

run_voting_multi

Run voting mode with multiple LLM providers