The Complete Production SOP: How to Deploy and Run Ollama Local Model Runner & Modelfile Config in Autonomous AI Agents

---
name: "ollama-local-model-runner"
description: "Run and customize open-source language models locally on CPU and Apple Silicon."
metadata:
  author: "Ollama"
  version: "1.0.0"
  runtime_compatibility:
    - "codex"
    - "claude-code"
    - "cursor"
    - "windsurf"
triggers:
  - "run model with ollama"
  - "create custom ollama modelfile"
  - "call ollama api python"
permissions:
  fs_read: true
  fs_write: "workspace"
  shell_exec: false
  net_access: false
---

FROM llama3.2:3b

# Set strict memory context and temperature
PARAMETER temperature 0.2
PARAMETER num_ctx 4096
PARAMETER stop "<|eot_id|>"

# Configure authoritative system directive
SYSTEM """
You are an expert deterministic code review assistant. Output pure valid JSON responses without narrative fluff.
"""

import json
import urllib.request

def call_ollama(prompt: str, model: str = "json-reviewer") -> dict:
    url = "http://127.0.0.1:11434/api/chat"
    payload = {
        "model": model,
        "messages": [{"role": "user", "content": prompt}],
        "stream": False,
        "format": "json"
    }
    req = urllib.request.Request(
        url,
        data=json.dumps(payload).encode("utf-8"),
        headers={"Content-Type": "application/json"}
    )
    with urllib.request.urlopen(req) as response:
        result = json.loads(response.read().decode("utf-8"))
        return json.loads(result["message"]["content"])

if __name__ == "__main__":
    print(call_ollama("Review: return x == y;"))

#!/usr/bin/env bash
set -euo pipefail

MODEL_NAME="json-reviewer"

# 1. Check if model exists in Ollama
if ! ollama list | grep -q "$MODEL_NAME"; then
  echo "Error: Model $MODEL_NAME not registered." >&2
  exit 1
fi

# 2. Test inference sanity check
RESPONSE=$(curl -fsSL -X POST http://127.0.0.1:11434/api/generate \
  -d "{\"model\": \"$MODEL_NAME\", \"prompt\": \"ping\", \"stream\": false}")

echo "Verification successful: Model responds cleanly."