Lesson 4.2 — Talk to Your Model from Code (Local API)

Introduction to AI · Module 4: Local LLMs · About 60 min · Windows PC, PowerShell 5, Ollama 0.40.11

Goal: call your local model from PowerShell and from Python, and build a summarise.py that works with the internet off.

Before you start

• Open PowerShell and run cd C:\llm-test
• Your 4.1 winner: (the examples use qwen3.8:27b; swap in your winner if it's different)

Step 1 · Is the server up? (⏱ 0:00–0:05)

Invoke-RestMethod http://localhost:11434/api/version

You should see:

version
-------
0.40.11

• Version shown (if it errors, start Ollama from the Start menu)

Step 2 · Chat via the API from PowerShell (⏱ 0:05–0:15)

Paste the whole block:

$body = @{
  model    = "qwen3.8:27b"
  messages = @(@{ role = "user"; content = "Give me 3 tips for focus, one line each." })
  stream   = $false
} | ConvertTo-Json -Depth 5
$bytes = [System.Text.Encoding]::UTF8.GetBytes($body)
$r = Invoke-RestMethod http://localhost:11434/api/chat -Method Post -Body $bytes -ContentType "application/json"
$r.message.content

You should see 3 one-line tips (after a short wait while the model loads).

• Got the tips

Step 3 · Check Python (⏱ 0:15–0:20)

python --version

You should see Python 3.x.x. If you get "not recognized", or the Microsoft Store opens:

winget install Python.Python.3.12

Then open a new PowerShell window, run cd C:\llm-test, and check again.

• Python 3 works

Step 4 · Build summarise.py (⏱ 0:20–0:35)

This uses only Python's built-in modules, so there's nothing extra to install. Paste the whole block:

$code = @'
import json, sys, urllib.request
MODEL = "qwen3.8:27b"
text = open(sys.argv[1], encoding="utf-8-sig").read()
req = {"model": MODEL, "stream": False, "messages": [
  {"role": "system", "content": "You summarise documents in 5 short bullet points, then list action items."},
  {"role": "user", "content": text}]}
data = json.dumps(req).encode("utf-8")
r = urllib.request.Request("http://localhost:11434/api/chat", data,
                           {"Content-Type": "application/json"})
print(json.load(urllib.request.urlopen(r))["message"]["content"])
'@
[IO.File]::WriteAllText("C:\llm-test\summarise.py", $code)
python summarise.py t2.txt

You should see 5 bullets and the action items for your task-2 notes from Lesson 4.1.

• summarise.py works

Step 5 · The OpenAI-compatible endpoint (⏱ 0:35–0:45)

Many apps speak "OpenAI". Ollama can pretend to be OpenAI, so those apps can use your local model.

python -m pip install openai
$code = @'
from openai import OpenAI
c = OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")
r = c.chat.completions.create(model="qwen3.8:27b",
      messages=[{"role": "user", "content": "Say hello in 5 words."}])
print(r.choices[0].message.content)
'@
[IO.File]::WriteAllText("C:\llm-test\openai_test.py", $code)
python openai_test.py

You should see a 5-word hello. The api_key can be any text; Ollama ignores it.

• OpenAI-style call works

Step 6 · The offline test (⏱ 0:45–0:55)

1. Turn off Wi-Fi and unplug Ethernet (or click the network icon → Airplane mode).
2. Check you're really offline:

Test-Connection 8.8.8.8 -Count 1 -Quiet

You should see False.

1. Run:

python summarise.py t2.txt

You should see the summary again, so it worked with no internet. Turn the network back on afterwards.

• Summary worked offline

✅ Pass check

• Step 2 PowerShell API call returned an answer
• python summarise.py t2.txt printed a summary with the internet off
• Told Skills Coach so 4.2 gets marked done

Troubleshooting