Back to Three Intelligence Pillars

Agentic Intelligence

Developer guide for configuring and using Agentic Intelligence—multi-agent coordination for intelligent strategy selection.

Overview

Agentic Intelligence uses a multi-agent system where research agents, validation engineers, and performance analysts collaborate to select the optimal optimization strategy for your problem. The system requires consensus (default 67%) before a strategy is approved.

How It Works

  1. 1.Research Agent analyzes your problem and suggests strategies based on literature and best practices
  2. 2.Validation Engineer tests strategies against constraints and risk models
  3. 3.Performance Analyst reviews historical data to predict performance
  4. 4.Consensus Engine requires agreement before strategy is approved

Simple Configuration

Enable Agentic Intelligence with a simple boolean flag:

Enable Agentic Intelligence
from sematryx import Sematryx

client = Sematryx(api_key="sk-your-api-key")

# Enable Agentic Intelligence
result = client.optimize(
    objective="minimize",
    variables=[{"name": "x", "bounds": (-5, 5)}, {"name": "y", "bounds": (-5, 5)}],
    objective_function=sphere,
    use_agentic_intelligence=True
)

Advanced Configuration

Fine-tune Agentic Intelligence behavior with advanced options:

Advanced Agentic Configuration
from sematryx import Sematryx

client = Sematryx(api_key="sk-your-api-key")

# Advanced Agentic Intelligence configuration
result = client.optimize(
    objective="minimize",
    variables=[{"name": "x", "bounds": (-5, 5)}, {"name": "y", "bounds": (-5, 5)}],
    objective_function=sphere,
    intelligence_config={
        "use_agentic_intelligence": True,
        "agentic": {
            "max_agents_per_problem": 5,  # Maximum number of agents (default: 3)
            "consensus_threshold": 0.67,  # Agreement threshold for strategy selection
            "agent_timeout": 30  # Timeout for agent responses in seconds
        }
    }
)

Configuration Options

  • max_agents_per_problem (int, default: 3)

    Maximum number of agents that will collaborate on strategy selection. More agents provide better consensus but increase latency.

  • consensus_threshold (float, default: 0.67)

    Minimum agreement percentage required before a strategy is approved. Range: 0.5 to 1.0. Higher values require more agreement but may be more conservative.

  • agent_timeout (int, default: 30)

    Timeout in seconds for agent responses. If an agent doesn't respond within this time, it's excluded from consensus.

REST API Configuration

Configure Agentic Intelligence via REST API:

REST API - Agentic Intelligence
curl -X POST https://api.sematryx.com/v1/optimize \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "objective_function_id": "func_1234567890",
    "bounds": [[-10, 10], [-10, 10]],
    "max_evaluations": 2000,
    "intelligence_config": {
      "use_agentic_intelligence": true,
      "agentic": {
        "max_agents_per_problem": 5,
        "consensus_threshold": 0.67,
        "agent_timeout": 30
      }
    }
  }'

JavaScript SDK Configuration

Configure Agentic Intelligence using the JavaScript SDK:

JavaScript SDK - Agentic Intelligence
import { Sematryx } from '@sematryx/javascript-sdk'

const client = new Sematryx('sk-your-api-key')

// Enable Agentic Intelligence
const result = await client.optimize({
  objective: 'minimize',
  variables: [
    { name: 'x', bounds: [-5, 5] },
    { name: 'y', bounds: [-5, 5] }
  ],
  objectiveFunction: sphere,
  intelligenceConfig: {
    useAgenticIntelligence: true,
    agentic: {
      maxAgentsPerProblem: 5,
      consensusThreshold: 0.67,
      agentTimeout: 30
    }
  }
})

Best Practices

  • •Use for complex problems: Agentic Intelligence is most valuable when you're unsure which algorithm to use or when problems have unusual characteristics.
  • •Balance agents vs latency: More agents provide better consensus but increase decision time. Use 3-5 agents for most cases.
  • •Adjust consensus threshold: Higher thresholds (0.75+) are more conservative but may reject valid strategies. Lower thresholds (0.6) are faster but less rigorous.
  • •Monitor agent timeouts: If agents frequently timeout, consider increasing the timeout value or reducing the number of agents.