How to Run 15 AI Agents on a $5/Month VPS (Complete Guide)

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Create systemd service with CPU optimization
sudo tee /etc/systemd/system/ollama.service << 'EOF'
[Unit]
Description=Ollama Service
After=network.target

[Service]
Type=simple
User=ollama
Environment="OLLAMA_NUM_THREAD=4"
Environment="OLLAMA_NUM_BATCH=1"
Environment="OLLAMA_NUM_PARALLEL=1"
Environment="OLLAMA_MAX_LOADED_MODELS=1"
Environment="OLLAMA_KEEP_ALIVE=24h"
Environment="OLLAMA_FLASH_ATTENTION=1"
Environment="OLLAMA_KV_CACHE_TYPE=q8_0"
ExecStart=/usr/local/bin/ollama serve
Restart=always
RestartSec=3

[Install]
WantedBy=multi-user.target
EOF

sudo systemctl enable ollama
sudo systemctl start ollama

ollama pull qwen3-abliterated:0.6b    # 396MB — instant on CPU
ollama pull qwen3-abliterated:1.7b    # 1.1GB — 25-40 tok/s
ollama pull deepseek-r1-abliterated:1.5b  # 1.1GB — reasoning
ollama pull tinyllama:1.1b            # 637MB — simple tasks
ollama pull phi4-mini                  # 2.3GB — code, 74.4% HumanEval

# api_chain.py — Production-tested fallback chain
import time, requests, json

class APIChain:
    def __init__(self):
        self.providers = [
            {"name": "groq", "func": self._groq, "min_interval": 4},
            {"name": "gemini", "func": self._gemini, "min_interval": 4},
            {"name": "openrouter", "func": self._openrouter, "min_interval": 2},
            {"name": "ollama", "func": self._ollama, "min_interval": 0},
        ]
        self.last_call = {}

    def chat(self, prompt, max_tokens=500, system="You are a helpful assistant."):
        for provider in self.providers:
            name = provider["name"]
            # Rate limit check
            if name in self.last_call:
                elapsed = time.time() - self.last_call[name]
                if elapsed < provider["min_interval"]:
                    time.sleep(provider["min_interval"] - elapsed)

            try:
                result = provider["func"](prompt, max_tokens, system)
                if result:
                    self.last_call[name] = time.time()
                    return result
            except Exception as e:
                print(f"{name} failed: {e}")
                continue

        return None  # All providers failed

    def _groq(self, prompt, max_tokens, system):
        from groq import Groq
        client = Groq(api_key=os.getenv("GROQ_API_KEY"))
        resp = client.chat.completions.create(
            model="llama-3.3-70b-versatile",
            messages=[{"role": "system", "content": system},
                      {"role": "user", "content": prompt}],
            max_tokens=max_tokens
        )
        content = resp.choices[0].message.content
        if not content:  # Reasoning models
            content = resp.choices[0].message.model_extra.get("reasoning", "")
        return content

    def _gemini(self, prompt, max_tokens, system):
        import google.generativeai as genai
        genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))
        model = genai.GenerativeModel("gemini-3.1-flash-lite")
        resp = model.generate_content(f"{system}\n\n{prompt}")
        return resp.text

    def _openrouter(self, prompt, max_tokens, system):
        resp = requests.post(
            "https://openrouter.ai/api/v1/chat/completions",
            headers={"Authorization": f"Bearer {os.getenv('OPENROUTER_API_KEY')}"},
            json={
                "model": "nvidia/nemotron-3.5-lightning:free",
                "messages": [{"role": "system", "content": system},
                             {"role": "user", "content": prompt}],
                "max_tokens": max_tokens
            }
        )
        return resp.json()["choices"][0]["message"]["content"]

    def _ollama(self, prompt, max_tokens, system):
        resp = requests.post(
            "http://localhost:11434/api/generate",
            json={
                "model": "qwen3-abliterated:1.7b",
                "prompt": f"{system}\n\n{prompt}",
                "stream": False,
                "options": {"num_ctx": 512, "num_predict": max_tokens, "temperature": 0.3}
            },
            timeout=120
        )
        return resp.json()["response"]

# agent_template.py — Base agent pattern
import time, json, os

class BaseAgent:
    def __init__(self, name, interval, api_chain):
        self.name = name
        self.interval = interval  # seconds between cycles
        self.api = api_chain
        self.results_file = f"memory/{name}_results.json"

    def run_forever(self):
        while True:
            try:
                task = self.get_task()
                if task:
                    result = self.api.chat(task["prompt"], max_tokens=500)
                    self.save_result(task, result)
                    self.log(f"Task completed: {task['name']}")
            except Exception as e:
                self.log(f"Error: {e}")

            time.sleep(self.interval)

    def get_task(self):
        # Override in subclass
        raise NotImplementedError

    def save_result(self, task, result):
        results = []
        if os.path.exists(self.results_file):
            with open(self.results_file) as f:
                results = json.load(f)
        results.append({
            "task": task["name"],
            "result": result,
            "timestamp": time.strftime("%Y-%m-%d %H:%M:%S")
        })
        with open(self.results_file, "w") as f:
            json.dump(results, f, indent=2)

    def log(self, msg):
        print(f"[{self.name}] {msg}")

# Create a deployment script
#!/bin/bash

# Agent definitions
agents=(
    "revenue_army:python3 agents/revenue_army.py:600"
    "shadow_council:python3 agents/shadow_council.py:900"
    "local_agent:python3 agents/local_agent.py:600"
    "agent_v2:python3 agents/agent_v2.py:900"
    "worker_v5d:python3 agents/worker.py:600"
    "revexec:python3 agents/revexec.py:900"
    "daemon_monitor:python3 agents/daemon.py:300"
    "vuln_scanner:python3 agents/vuln_scanner.py:3600"
    "airdrop_ninja:python3 agents/airdrop.py:3600"
    "m2m_tracker:python3 agents/m2m.py:300"
    "algo_trader:python3 agents/trader.py:60"
    "x402_server:python3 agents/x402.py:0"
    "freelance_hunter:python3 agents/freelance.py:1800"
    "data_nexus:python3 agents/datanexus.py:0"
    "flash_arb:python3 agents/arb.py:60"
)

for agent_def in "${agents[@]}"; do
    IFS=':' read -r name cmd interval <<< "$agent_def"

    # Check if screen already exists
    if ! screen -ls | grep -q "$name"; then
        screen -dmS "$name" bash -c "$cmd"
        echo "Started: $name"
    else
        echo "Already running: $name"
    fi
done

echo "All agents deployed."

# daemon_monitor.py — Checks all agents every 5 minutes
import subprocess, json, time

AGENTS = [
    "revenue_army", "shadow_council", "local_agent",
    "agent_v2", "worker_v5d", "revexec",
    "daemon_monitor", "vuln_scanner", "airdrop_ninja",
    "m2m_tracker", "algo_trader", "x402_server",
    "freelance_hunter", "data_nexus", "flash_arb"
]

COMMANDS = {
    "revenue_army": "python3 agents/revenue_army.py",
    "shadow_council": "python3 agents/shadow_council.py",
    # ... (map each agent to its command)
}

def check_and_restart():
    while True:
        screens = subprocess.run(
            ["screen", "-ls"], capture_output=True, text=True
        ).stdout

        for agent in AGENTS:
            if agent not in screens:
                cmd = COMMANDS.get(agent, "")
                if cmd:
                    subprocess.run(
                        ["screen", "-dmS", agent, "bash", "-c", cmd],
                        capture_output=True
                    )
                    log(f"Restarted: {agent}")

        time.sleep(300)  # Check every 5 minutes

def log(msg):
    print(f"[DAEMON] {time.strftime('%H:%M:%S')} {msg}")

if __name__ == "__main__":
    check_and_restart()

# /etc/environment — CPU optimization
OLLAMA_NUM_THREAD=4              # Match physical cores
OLLAMA_NUM_BATCH=1               # Single-sequence for CPU
OLLAMA_NUM_PARALLEL=1            # One request at a time
OLLAMA_MAX_LOADED_MODELS=1       # Never load 2 models on CPU
OLLAMA_KEEP_ALIVE=24h            # Keep model in RAM
OLLAMA_FLASH_ATTENTION=1         # Speed boost
OLLAMA_KV_CACHE_TYPE=q8_0        # Save RAM

# Monitor CPU load
watch -n 5 uptime
# If load > 10: kill non-essential screens
# If load < 5: Ollama works perfectly

# Check CPU load
uptime
# If load > 10, kill non-essential screens
screen -ls
screen -X -S non_essential quit

# Add delay and retry
if response.status_code == 429:
    time.sleep(60)  # Wait 1 minute
    # Try next provider in chain

# Daemon auto-restarts, but manual check:
screen -ls
# If missing:
screen -dmS agent_name bash -c "python3 agents/agent.py"

# Check RAM
free -h
# If swap is heavy, reduce Ollama models
ollama rm phi4-mini  # Remove largest model