lms --help

~/.lmstudio/models/
└── publisher/
    └── model/
        └── model-file.gguf

Open LM Studio
→ Search model
→ Download small model
→ Load model
→ Chat
→ Open Developer tab
→ Start local server

Qwen 2.5 / Qwen 3 small instruct models
Llama 3.2 1B/3B
Gemma small models
Mistral 7B if enough RAM/VRAM

# Start daemon
lms daemon up

# Search/download model
lms get qwen

# List downloaded models
lms ls

# Load model
lms load <model-name> --gpu=auto --context-length=4096

# See loaded models
lms ps

# Terminal chat
lms chat

# Start API server
lms server start

# Stop API server
lms server stop

lms server start

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:1234/v1",
    api_key="lm-studio"
)

response = client.chat.completions.create(
    model="local-model",
    messages=[
        {"role": "user", "content": "Explain RAG with one example."}
    ],
    temperature=0.2,
)

print(response.choices[0].message.content)

1. Download a small instruct model.
2. Attach a PDF.
3. Ask: “Summarize section 2 only.”
4. Ask: “Give exact bullet points from the document.”
5. Ask: “Which page or section supports this?”

- Quickly testing many models
- Teaching local LLMs visually
- Comparing GGUF vs MLX models
- Running a local OpenAI-compatible endpoint
- Document chat demos
- Testing prompt templates
- Testing structured output and tool use
- Using MCP tools locally

# Remove app using OS method first.
# Optional: delete downloaded models and LM Studio local data.
rm -rf ~/.lmstudio

# Uninstall from Windows Apps first.
# Optional: delete local data and models.
Remove-Item -Recurse -Force "$env:USERPROFILE\.lmstudio"

- Best beginner experience
- GUI model search/download
- Local API server
- Good for demos and teaching
- Supports document chat
- Supports llama.cpp and MLX runtimes

- Desktop app is proprietary
- Not the best serious server deployment tool
- Less reproducible than pure Docker/server setups

http://localhost:11434/api

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

irm https://ollama.com/install.ps1 | iex

docker run -d \
  --name ollama \
  -p 11434:11434 \
  -v ollama:/root/.ollama \
  ollama/ollama

Linux service files  → /etc/systemd/system/ollama.service
Linux models/data    → /usr/share/ollama
macOS app            → /Applications/Ollama.app
macOS CLI symlink    → /usr/local/bin/ollama
macOS user data      → ~/.ollama and ~/Library/... Ollama folders
Windows              → app install + user .ollama directory

ollama run llama3.2:3b

ollama list                 # list downloaded models
ollama ps                   # list running models
ollama pull llama3.2:3b     # download model
ollama run llama3.2:3b      # run model
ollama rm llama3.2:3b       # remove model
ollama show llama3.2:3b     # model info
ollama serve                # start server manually

curl http://localhost:11434/api/generate \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama3.2:3b",
    "prompt": "Explain Docker in simple words.",
    "stream": false
  }'

curl http://localhost:11434/api/chat \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama3.2:3b",
    "messages": [
      {"role": "system", "content": "You are a helpful Linux tutor."},
      {"role": "user", "content": "Explain chmod 755."}
    ],
    "stream": false
  }'

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama"
)

response = client.chat.completions.create(
    model="llama3.2:3b",
    messages=[
        {"role": "user", "content": "Write a beginner-friendly Git explanation."}
    ],
)

print(response.choices[0].message.content)

FROM llama3.2:3b

SYSTEM """
You are a patient Python tutor.
Use small examples.
Ask one practice question at the end.
"""

PARAMETER temperature 0.2
PARAMETER num_ctx 8192

ollama create python-tutor -f Modelfile
ollama run python-tutor

ollama pull nomic-embed-text

curl http://localhost:11434/api/embed \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nomic-embed-text",
    "input": "Local LLMs can run on your laptop."
  }'

PDF search
document Q&A
semantic search
local knowledge base

curl http://localhost:11434/api/generate \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama3.2:3b",
    "prompt": "Extract name and skills from: Riya knows Python and Docker.",
    "format": "json",
    "stream": false
  }'

- Calculator tool
- Weather lookup
- Database query
- File search
- Coding agent command execution

Same prompt
Same context length
Same temperature
Compare tokens/sec and answer quality

ollama launch codex

sudo systemctl stop ollama
sudo systemctl disable ollama
sudo rm -f /etc/systemd/system/ollama.service

# Remove binary
sudo rm -f "$(which ollama)"

# Remove downloaded models/data
sudo rm -rf /usr/share/ollama ~/.ollama

# Remove service user/group
sudo userdel ollama 2>/dev/null || true
sudo groupdel ollama 2>/dev/null || true

sudo rm -rf /Applications/Ollama.app
sudo rm -f /usr/local/bin/ollama
rm -rf ~/.ollama
rm -rf ~/Library/Application\ Support/Ollama
rm -rf ~/Library/Caches/ollama
rm -rf ~/Library/Caches/com.electron.ollama
rm -rf ~/Library/WebKit/com.electron.ollama

- Very easy CLI
- Great local API
- Good for prototypes
- Good for coding tool integrations
- Official Docker image
- Model management is simple

- Less low-level tuning than llama.cpp
- Not the strongest multi-user production server
- Model abstraction hides some internals

[ ] I know how to install and start LM Studio.
[ ] I can use the `lms` CLI to download and manage models.
[ ] I can start the LM Studio local API server.
[ ] I know how to pull and run models with Ollama (`ollama run`).
[ ] I can use the Ollama REST API to generate text and chat.
[ ] I understand how to create a custom `Modelfile` in Ollama.