Heavy Swarm Example

export SWARMS_API_KEY="your-api-key-here"

import requests
import json
import os

API_BASE_URL = "https://api.swarms.world"
API_KEY = os.environ.get("SWARMS_API_KEY", "your_api_key_here")

headers = {
    "x-api-key": API_KEY,
    "Content-Type": "application/json"
}

def run_heavy_analysis(task: str, heavy_swarm_max_loops: int = 1, max_loops: int = 1) -> dict:
    """Run a deep multi-perspective analysis using HeavySwarm."""

    swarm_config = {
        "name": "Market Intelligence Swarm",
        "description": "Deep multi-agent research and analysis",
        "swarm_type": "HeavySwarm",
        "task": task,
        "agents": [],
        "heavy_swarm_max_loops": heavy_swarm_max_loops,
        "heavy_swarm_question_agent_model_name": "gpt-4.1",
        "heavy_swarm_worker_model_name": "claude-sonnet-4-20250514",
        "max_loops": max_loops
    }

    response = requests.post(
        f"{API_BASE_URL}/v1/swarm/completions",
        headers=headers,
        json=swarm_config,
        timeout=300
    )

    return response.json()

# Define a complex research task
task = """
Analyze the competitive landscape of the AI infrastructure market in 2025.
Focus on: cloud GPU providers, inference optimization startups, open-source
model serving frameworks, and edge AI deployment platforms. Evaluate market
sizing, key players, moats, and identify the most promising investment
opportunities with risk-adjusted return potential.
"""

# Run HeavySwarm analysis
result = run_heavy_analysis(task)

# Display results from each specialized agent
for output in result.get("output", []):
    agent = output["role"]
    content = output["content"]

    print(f"\n{'='*60}")
    print(f"{agent}")
    print(f"{'='*60}")

    # Handle content that may be a dict (question generator) or string
    if isinstance(content, dict):
        for key, value in content.items():
            if key != "thinking":
                print(f"\n{key}:")
                print(f"  {value}")
    else:
        print(str(content)[:800] + "...")

============================================================
Question Generator Agent
============================================================

research_question:
  What are the current market sizes, growth rates, and key players across cloud GPU,
  inference optimization, model serving, and edge AI segments?

analysis_question:
  What statistical patterns emerge from funding rounds, revenue multiples, and
  customer adoption rates across AI infrastructure subsectors?

alternatives_question:
  What are the highest risk-adjusted investment strategies considering direct equity,
  infrastructure ETFs, and picks-and-shovels approaches across these segments?

verification_question:
  How do reported market projections and competitive moat claims align with verified
  deployment data, customer churn rates, and technical benchmarks?

============================================================
Research-Agent
============================================================
## AI Infrastructure Market Research

### Market Overview
The global AI infrastructure market reached approximately $65B in 2024 and is
projected to exceed $120B by 2027 (38% CAGR).

### Cloud GPU Providers
Key players: NVIDIA (dominant), AMD (growing share), AWS Trainium/Inferentia
(custom silicon), Google TPU, CoreWeave (GPU-as-a-service)

Market dynamics: Severe GPU shortage easing in 2025, but demand continues to
outpace supply for H100/B200 clusters...

============================================================
Analysis-Agent
============================================================
## Statistical Analysis of AI Infrastructure Trends

### Funding Pattern Analysis
- Total VC funding in AI infra: $18.2B in 2024 (up 142% from 2023)
- Median Series B valuation: 45x ARR for inference optimization startups
- Customer acquisition efficiency: Cloud GPU providers show 0.8x CAC/LTV
  ratio vs 1.2x for edge AI platforms

### Adoption Rate Analysis
- Enterprise GPU cloud adoption: 67% of Fortune 500 (up from 34% in 2023)
- Open-source model serving: vLLM and TensorRT-LLM dominating with 78%
  combined market share in inference workloads...

============================================================
Alternatives-Agent
============================================================
## Investment Strategy Alternatives

### Strategy 1: Infrastructure Picks-and-Shovels (Recommended)
Focus: NVIDIA, networking (Arista, Broadcom), power/cooling
Risk: Moderate | Expected Return: 25-35% annualized
Rationale: Benefits from all AI growth regardless of which models win

### Strategy 2: Pure-Play Inference Optimization
Focus: Emerging startups (Groq, Cerebras, Together AI)
Risk: High | Expected Return: 3-10x over 5 years (venture-style)
Rationale: Inference costs are the primary bottleneck...

============================================================
Verification-Agent
============================================================
## Verification Assessment

### Claim: AI infra market at $65B (2024)
Status: VERIFIED
Sources: Gartner ($62-68B range), IDC ($64.5B), cross-referenced with
public company revenues (NVIDIA data center: $47.5B alone)

### Claim: 38% CAGR through 2027
Status: PARTIALLY VERIFIED
Note: Estimates range from 28-45% depending on methodology. Conservative
base case of 30% is better supported by deployment data...

============================================================
Synthesis-Agent
============================================================
## Executive Summary

The AI infrastructure market represents a verified $65B opportunity growing
at 30-38% CAGR. Our multi-agent analysis reveals strong consensus on three
key findings:

1. **Infrastructure layer offers best risk-adjusted returns** - All four
   specialist agents converge on picks-and-shovels as the safest approach
2. **Inference optimization is the highest-growth subsector** - 142% YoY
   funding growth with verified demand signals
3. **Edge AI remains early-stage** - Higher risk but strategic importance
   for the 2026-2028 cycle...

# Run a deeper 2-loop analysis for thorough due diligence
deep_result = run_heavy_analysis(
    task="Conduct due diligence on CoreWeave as a potential investment. Evaluate their GPU cloud infrastructure business model, competitive positioning against AWS/Azure/GCP, financial health, customer concentration risk, and long-term defensibility.",
    heavy_swarm_max_loops=2,
    max_loops=2
)

# The synthesis agent output from the final loop contains the most refined analysis
outputs = deep_result.get("output", [])
for output in outputs:
    if output["role"] == "Synthesis-Agent":
        print("FINAL SYNTHESIZED ANALYSIS:")
        print(str(output["content"])[:2000])