Five Ways Your Agents Collaborate
Swarm, pipeline, debate, crowd simulate, or specialize. Five coordination modes for any trading scenario. Monitor it all in real-time with the God's Eye force-directed fleet visualization.
Independent Agents, Maximum Coverage
In Swarm mode, each agent operates independently, scanning different markets or applying different strategies simultaneously. No consensus required -- every agent trades on its own signals, giving you maximum market coverage with diversified approaches.
Swarm mode is ideal when you want broad exposure across many markets without the overhead of inter-agent communication. Each agent has its own bankroll allocation, risk limits, and performance tracking.
- Best for: Broad market coverage across categories
- Risk profile: Diversified, uncorrelated strategies
- Latency: Lowest -- no inter-agent communication
"mode": "swarm",
"agents": [
"politics_momentum",
"sports_sentiment",
"weather_contrarian",
"econ_arbitrage"
],
"allocation": "equal",
"shared_risk_limit": false
}
Weighted Voting for High-Conviction Trades
Council mode aggregates signals from multiple agents using weighted voting. Each agent casts a confidence-weighted vote, and the system computes a weighted consensus before executing. Agents with better track records automatically earn higher weights.
This is ideal for high-stakes trades where you want multiple perspectives but need a single, decisive action. The weighting system ensures that historically accurate agents have more influence over the final decision.
- Best for: High-conviction, single-market decisions
- Risk profile: Consensus reduces false signals
- Weights: Auto-adjusted by historical accuracy
"mode": "council",
"min_consensus": 0.65,
"weight_method": "performance",
"weight_window": "30d",
"min_voters": 3,
"tie_breaker": "abstain"
}
Sequential Veto Chain for Maximum Safety
Pipeline mode passes trade signals through a sequential chain of agents. Each agent in the chain can approve or veto the trade. If any agent vetoes, the trade is killed immediately. Only trades that survive every gate get executed.
Think of it as a series of increasingly strict filters. The first agent generates a signal, the second checks fundamentals, the third validates risk parameters, and the fourth confirms timing. This produces a stricter review chain before execution.
- Best for: High-stakes trades requiring maximum validation
- Risk profile: Strict sequential validation
- Trade-off: Fewer trades, but higher conviction
"mode": "pipeline",
"chain": [
"signal_gen",
"fundamental_check",
"risk_gate",
"timing_confirm"
],
"veto_is_final": true,
"timeout_per_stage": "5s"
}
Reason: "Position size exceeds current volatility-adjusted limit. Daily VaR already at 82% of max."
AI-to-AI Debate for Nuanced Decisions
Deliberation mode enables structured debate between agents. Each agent presents its analysis, challenges other agents' reasoning, and can update its position based on new arguments. The coordinator synthesizes the final consensus after a configurable number of rounds.
This produces the most nuanced analysis, surfacing edge cases and counterarguments that any single agent might miss. Agents can change their votes mid-deliberation as they encounter persuasive reasoning from peers.
- Best for: Complex, nuanced market questions
- Risk profile: Highest-quality analysis
- Trade-off: Higher latency, deeper reasoning
"mode": "deliberation",
"max_rounds": 5,
"min_consensus": 0.70,
"allow_vote_change": true,
"early_stop": true,
"log_transcript": true
}
10,000 Virtual Traders, One Consensus
Spawn a crowd of up to 10,000 simulated market participants across 7 behavioral archetypes (Retail, Sharp, Political, Arb Bot, Hedger, Social, Contrarian) on a Watts-Strogatz social graph. Three robustness tiers (Scout, Analyst, Strategist) control simulation depth with ensemble averaging.
- Predict how the crowd reprices a contract — not the outcome itself
- LLM interprets consensus vs. current price for a directional signal
- Available as a coordination mode or standalone in the Lab
"mode": "crowd_sim",
"participants": 5000,
"robustness": "analyst",
"hours": 6,
"ensemble_runs": 3
}
See Your Entire Fleet in Real-Time
God's Eye is a force-directed particle visualization of your agent fleet. Each agent is a node, colored by state (green = running, amber = paused, red = error), sized by PnL contribution. Signal flow between team members renders as animated particle streams.
Choose the Right Mode for the Job
Each mode serves a different purpose. Mix and match across your portfolio.
Coordinate Your Agent Fleet
One agent is a tool. A coordinated team is an edge. Build yours in minutes.