Skip to main content
The Gnosis Prediction Market Agent platform includes over 60 specialized agents, each designed with unique trading strategies, models, and capabilities. These agents range from simple random traders to sophisticated AI-powered prediction systems.

​
Agent Categories

Agents are organized into the following categories:

Simple Agents

Basic agents including coinflip and known outcome predictors

Research Agents

Advanced agents with web research and analysis capabilities

Prophet Agents

LLM-powered agents using GPT-4, Claude, Gemini, and more

Microchain Agents

Self-learning agents with modifiable prompts and goal management

Specialized Agents

Purpose-built agents for arbitrage, social media, and market monitoring

​
Running Agents

All agents in the gallery can be run using the run_agent.py entrypoint: 
python prediction_market_agent/run_agent.py <agent_name> <market_type>

​
Example

# Run the GPT-4o Prophet agent on Omen markets
python prediction_market_agent/run_agent.py prophet_gpt4o omen

# Run the coinflip agent on Manifold markets
python prediction_market_agent/run_agent.py coinflip manifold

​
Agent Architecture

All agents inherit from the DeployableAgent base class from the prediction-market-agent-tooling library. This provides:
  • Market fetching and filtering
  • Trade execution and position management
  • Betting strategy configuration
  • Monitoring and logging
  • Deployment utilities for GKE

​
Key Agent Methods

Agents implement these core methods: 
class DeployableAgent:
    def load(self) -> None:
        """Initialize agent resources and models"""
        
    def answer_binary_market(self, market: AgentMarket) -> ProbabilisticAnswer | None:
        """Generate prediction for binary markets"""
        
    def verify_market(self, market_type: MarketType, market: AgentMarket) -> bool:
        """Filter markets to trade on"""
        
    def get_betting_strategy(self, market: AgentMarket) -> BettingStrategy:
        """Configure bet sizing and risk management"""

​
Performance Tracking

Agent performance can be tracked using:
  • Langfuse: All agents support tracing via enable_langfuse=True
  • Database Storage: Predictions and trades are stored in PostgreSQL
  • Metrics: PnL, accuracy, and bet statistics available via the API

​
Adding New Agents

To add a new agent to the platform:
  1. Create your agent class inheriting from DeployableAgent
  2. Add it to the RunnableAgent enum in run_agent.py
  3. Register it in the RUNNABLE_AGENTS dictionary
  4. Deploy to GKE or run locally
from prediction_market_agent_tooling.deploy.agent import DeployableTraderAgent

class MyCustomAgent(DeployableTraderAgent):
    def answer_binary_market(self, market: AgentMarket) -> ProbabilisticAnswer | None:
        # Your prediction logic here
        return ProbabilisticAnswer(
            p_yes=Probability(0.7),
            confidence=0.8,
            reasoning="My analysis shows..."
        )

​
Next Steps

Simple Agents

Start with basic agents

Prophet Agents

Explore LLM-powered predictions

Build docs developers (and LLMs) love

Get started for freeTalk to us