AI agents

One script, agent.py, holds a pretend network of three servers, four tools that read or change it, and the agent loop. The tools are real Python functions, so their results, the step limit and the approval question always behave exactly as shown.

Which tool the model picks, and its wording, can differ a little on your laptop. On macOS or Linux, use python3 and nano.

1Ask without tools

Run this:

mkdir C:\ai-lab\agent
cd C:\ai-lab\agent
notepad agent.py

Paste this whole block into agent.py, then save:

import json
import urllib.request

MODEL = "llama3.2:3b"
QUESTION = "Nobody on the 2nd floor can print. Can you check?"
USE_TOOLS = False     # True = let the model call the tools below
APPROVE = True        # True = ask you before any tool that changes something
MAX_STEPS = 6         # stop after this many calls to the model

# A pretend network, so the lab is safe to run on any laptop
SERVERS = {
    "web01": {"role": "intranet web server", "disk": 41, "services": {"nginx": "running"}},
    "file01": {"role": "file server", "disk": 97, "services": {"smb": "running"}},
    "print01": {"role": "print server for floors 1 to 3", "disk": 58, "services": {"spooler": "stopped"}},
}


def list_servers():
    return "\n".join(f"{name}: {s['role']}" for name, s in SERVERS.items())


def check_disk(server):
    return f"{server}: disk {SERVERS[server]['disk']}% used"


def check_service(server, service):
    state = SERVERS[server]["services"].get(service, "not installed")
    return f"{service} on {server}: {state}"


def restart_service(server, service):
    SERVERS[server]["services"][service] = "running"
    return f"{service} on {server} restarted, now running"


FUNCS = {f.__name__: f for f in [list_servers, check_disk, check_service, restart_service]}
CHANGES = {"restart_service"}      # tools that change something


def tool(name, description, **params):
    props = {p: {"type": "string", "description": d} for p, d in params.items()}
    return {"type": "function", "function": {"name": name, "description": description,
            "parameters": {"type": "object", "properties": props, "required": list(params)}}}


TOOLS = [
    tool("list_servers", "List all servers and what each one does."),
    tool("check_disk", "Show how full a server's disk is.", server="Server name, for example web01"),
    tool("check_service", "Show whether a service is running.", server="Server name, for example web01", service="Service name, for example nginx"),
    tool("restart_service", "Restart a service. Only use it when check_service shows a problem.", server="Server name, for example web01", service="Service name, for example nginx"),
]


def chat(messages):
    body = {"model": MODEL, "messages": messages, "stream": False, "options": {"temperature": 0}}
    if USE_TOOLS:
        body["tools"] = TOOLS
    req = urllib.request.Request("http://localhost:11434/api/chat", data=json.dumps(body).encode(),
                                 headers={"Content-Type": "application/json"})
    with urllib.request.urlopen(req) as r:
        return json.load(r)


messages = [{"role": "system", "content": "You are an IT support agent. Check with the tools before you answer. Answer in two sentences."},
            {"role": "user", "content": QUESTION}]
print("Question:", QUESTION)
for step in range(1, MAX_STEPS + 1):
    msg = chat(messages)["message"]
    messages.append(msg)
    calls = msg.get("tool_calls")
    if not calls:
        print("Answer:", msg["content"].strip())
        break
    for call in calls:
        name, args = call["function"]["name"], call["function"]["arguments"]
        print(f"Step {step}: {name}(" + ", ".join(f"{k}={v!r}" for k, v in args.items()) + ")")
        if name in CHANGES and APPROVE and input("  Allow this change? [y/n] ").strip().lower() != "y":
            result = "The user did not allow this change."
        else:
            result = FUNCS[name](**args)
        print("  -> " + result.replace("\n", "\n     "))
        messages.append({"role": "tool", "content": result, "tool_name": name})
else:
    print(f"Stopped: no answer after {MAX_STEPS} steps.")

Look at SERVERS: the print spooler on print01 is stopped. That’s the fault the agent has to find. First, ask with tools switched off:

Run this:

python agent.py

You should see something like this (the model’s words and choices may differ a little):

Question: Nobody on the 2nd floor can print. Can you check?
Answer: It could be a paper jam, an empty tray, a driver problem or the print spooler service. Try restarting your computer and the printer, and if it still fails, contact your IT team.

One answer, no steps. It’s sensible advice, and it even mentions the spooler, but it’s a list of guesses. The model can’t see your servers.

✅ Checkpoint: You get an Answer line and no Step lines.

2Turn the tools on

Change this line:

USE_TOOLS = True

Run this:

python agent.py

When the script asks Allow this change?, type y and press Enter.

You should see something like this (the model’s words and choices may differ a little):

Question: Nobody on the 2nd floor can print. Can you check?
Step 1: list_servers()
  -> web01: intranet web server
     file01: file server
     print01: print server for floors 1 to 3
Step 2: check_service(server='print01', service='spooler')
  -> spooler on print01: stopped
Step 3: restart_service(server='print01', service='spooler')
  Allow this change? [y/n] y
  -> spooler on print01 restarted, now running
Answer: The print spooler on print01 had stopped, so nobody could print. I restarted it and it is running again, so printing should work now.

Nobody wrote “if printing fails, check the spooler”. The model chose each step from the tool descriptions and the results so far: list the servers, check the print server, restart the stopped service, then answer. The restart only ran because you typed y.

✅ Checkpoint: Three steps, an approval question, and an answer that says the spooler on print01 was restarted.

3Ask something different

Change this line:

QUESTION = "Is any server running out of disk space?"

Run this:

python agent.py

You should see something like this (the model’s words and choices may differ a little):

Question: Is any server running out of disk space?
Step 1: list_servers()
  -> web01: intranet web server
     file01: file server
     print01: print server for floors 1 to 3
Step 2: check_disk(server='web01')
  -> web01: disk 41% used
Step 3: check_disk(server='file01')
  -> file01: disk 97% used
Step 4: check_disk(server='print01')
  -> print01: disk 58% used
Answer: Yes: file01 is almost full at 97% used, so free up space or add storage soon. web01 (41%) and print01 (58%) are fine.

Different question, different plan: list the servers, then check each disk. Nothing in the code says how to answer a disk question; the loop is the same, only the model’s choices changed. That flexibility is the whole point of an agent, and also the reason it needs limits.

✅ Checkpoint: Four steps, and the answer names file01 at 97%.

4Set a step limit

Change this line:

MAX_STEPS = 3

Run this:

python agent.py

You should see something like this (the model’s words and choices may differ a little):

Question: Is any server running out of disk space?
Step 1: list_servers()
  -> web01: intranet web server
     file01: file server
     print01: print server for floors 1 to 3
Step 2: check_disk(server='web01')
  -> web01: disk 41% used
Step 3: check_disk(server='file01')
  -> file01: disk 97% used
Stopped: no answer after 3 steps.

The agent needed four steps and only got three, so your code stopped it. Each step is a full model call, and a confused agent can call the same tool again and again.

A step limit puts a ceiling on time and cost. Real systems also limit total tokens and money per task.

✅ Checkpoint: The run ends with Stopped: no answer after 3 steps.