Swarm Intelligence / War Room

The War Room is OpenZess's multi-agent collaboration system. It lets you deploy multiple AI agents simultaneously to debate, analyze, and synthesize solutions.

How It Works

Swarm Debate Mode

When you enable Swarm Mode in the chat, your message is sent to multiple AI agents in parallel. Each agent has a different role and provider:

• Strategist (DeepSeek) — Plans the high-level approach
• Critic (DeepSeek) — Finds flaws and challenges assumptions
• Optimizer (GLM) — Refines and optimizes the solution

The agents take turns debating in a structured conversation. If all agents agree, they output [CONSENSUS REACHED] and the debate ends.

The Final Synthesis

After the debate, a Final Judge (your primary model) reads the entire transcript and synthesizes a definitive answer as a structured Markdown table.

Architecture

class SwarmManager:
    def __init__(self):
        self.executor = concurrent.futures.ThreadPoolExecutor(max_workers=10)

    async def dispatch_squad_stream(self, prompt, squad_config):
        # Fire off all agents in parallel with 400ms stagger
        for i, config in enumerate(squad_config):
            delay = i * 400  # Prevents API rate limiting
            futures.append(
                loop.run_in_executor(self.executor, run_agent_stream, ...)
            )

Why 400ms Stagger?

Launching all API calls simultaneously can trigger rate limits and socket exhaustion. The 400ms delay between each agent launch provides just enough breathing room.

Squad Configuration

Each squad member is defined by:

{
    "role_name": "Strategist",
    "provider": "deepseek2",
    "api_key": "sk-...",
    "system_instruction": "You are a strategic planner..."
}

Use Cases

• Code Review: Let one agent write code while another critiques it
• Research: Deploy multiple agents to search different aspects of a topic
• Decision Making: Get multiple AI perspectives before making a choice
• Brainstorming: Let agents build on each other's ideas iteratively