How to Install and Run Local LLMs: Complete Setup Guide for Ollama & LM Studio

Running local LLMs in 2026 is easier and more powerful than ever—Ollama and LM Studio offer two of the fastest paths to offline, private AI on your own hardware. For Ollama, head to ollama.com/download, grab the installer for Windows/macOS or run the one-liner

Step-by-step tutorials for Windows, Mac, and Linux • No technical experience required

---

If you searched “install ollama,” “ollama windows,” or “lm studio download,” you’re in the right place. Setting up local AI tools is easier than you think—whether you prefer clicking buttons or typing commands.

This guide covers both Ollama and LM Studio across every major operating system. By the end, you’ll have a working local LLM on your machine, ready to chat, code, or integrate into projects.

---

📊 What Users Are Searching for Right Now

Search Query	Trend	What It Tells Us
install ollama	+7%	More people setting up Ollama
ollama run	+8%	Users moving from install to usage
ollama windows	High volume	Windows setup demand remains strong
ollama mac	+5%	Mac users adopting Ollama
download ollama	Steady	Consistent onboarding interest
how to use ollama	Steady	Beginner education gap

Key Insight: Users want platform-specific, actionable setup guides—not general overviews.

---

🖥️ Part 1: LM Studio Setup (The GUI Path)

Best for: Windows beginners, non-technical users, anyone who hates terminals

✅ System Requirements

• Windows: 10 or 11, 8GB+ RAM (16GB recommended)
• Mac: Intel or Apple Silicon, 8GB+ RAM
• Storage: 10GB+ free space (models are large!)
• GPU (optional but faster): NVIDIA CUDA, AMD, or Apple Metal

📦 Step 1: Download & Install

1. Visit lmstudio.ai
2. Click Download for [Your OS]
3. Run the installer (.exe for Windows, .dmg for Mac)
4. Drag to Applications folder (Mac) or follow installer wizard (Windows)
5. Launch LM Studio

Time: 2 minutes ⏱️

🎯 Step 2: Download Your First Model

1. Click the magnifying glass icon (Search)
2. Browse popular models or search for:
   • Llama 3.2 3B (fast, good for beginners)
   • Mistral 7B (balanced performance)
   • Phi-3 mini (great on low RAM)
3. Click Download on your chosen model
4. Wait for download to complete (varies by model size)

Pro tip: Start with smaller models (1B–7B parameters) to test performance.

💬 Step 3: Load and Chat

1. Go to AI Chat tab (left sidebar)
2. Select your downloaded model from dropdown
3. Wait for “Model loaded” indicator
4. Start typing! 🎉

Optional tweaks:

• Click gear icon → Adjust temperature, context length
• Enable GPU offloading in Settings → Hardware

🪟 Windows-Specific Tips

• DirectML support: Faster on AMD/Intel GPUs
• Task Manager: Monitor RAM usage (Ctrl+Shift+Esc)
• Dark mode: Settings → Appearance

🍏 Mac-Specific Tips

• Apple Silicon: MLX acceleration enabled by default
• Intel Macs: Still works, but slower
• Activity Monitor: Check memory pressure

---

⌨️ Part 2: Ollama Setup (The CLI Path)

Best for: Developers, Mac/Linux users, API seekers, Claude fans

✅ System Requirements

• Windows: 10/11, WSL2 optional but recommended
• Mac: Intel or Apple Silicon
• Linux: Most distros supported
• RAM: 8GB minimum, 16GB+ recommended
• GPU: Automatic CUDA/Metal/ROCm detection

📦 Step 1: Install Ollama

🍏 Mac (Easiest):

bash

brew install ollama

No Homebrew? Download from ollama.com

🪟 Windows:

1. Download OllamaSetup.exe from ollama.com
2. Run installer
3. Open PowerShell or Command Prompt
4. Verify installation:

bash

ollama --version

🐧 Linux:

bash

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

Time: 1–3 minutes ⏱️

🎯 Step 2: Download and Run Your First Model

One-line magic:

bash

ollama run llama3.2

That’s it. Ollama automatically:

1. Downloads the model (3B parameters)
2. Loads it into memory
3. Opens an interactive chat session

Other beginner-friendly models:

bash

ollama run phi3          # Microsoft's efficient model
ollama run mistral       # Strong all-rounder
ollama run qwen2.5:7b    # Growing +20% trend

💬 Step 3: Chat and Exit

Once the model loads:

text

>>> Hello, who are you?
I'm Llama, an AI assistant...

>>> Write a Python function to reverse a string
def reverse_string(s):
    return s[::-1]

>>> /exit   # or Ctrl+D

Pro tip: ollama run handles everything in one command. No separate download steps needed.

🖥️ Platform-Specific Ollama Optimizations

🍏 Mac (Apple Silicon):

• Metal GPU acceleration works automatically
• Check with: ollama run llama3.2 --verbose
• Homebrew installation strongly recommended

🪟 Windows:

• WSL2 method (best performance):

bash

# Inside WSL2 terminal
curl -fsSL https://ollama.com/install.sh | sh
ollama run llama3.2

• Native PowerShell: Works fine, GPU support via DirectML

🐧 Linux:

• Systemd service available: sudo systemctl enable ollama
• NVIDIA GPU: sudo apt install nvidia-cuda-toolkit
• AMD GPU: ROCm support

---

🔄 Part 3: Keeping Everything Updated

LM Studio Updates

• Automatic: App prompts when new version available
• Manual: Settings → Check for Updates
• Models: Re-download newer versions when released

Ollama Updates

bash

# Check current version
ollama --version

# Update Ollama itself
ollama update

# Update models (pull latest)
ollama pull llama3.2
ollama pull mistral

Search trend: ollama update sees regular interest—don’t skip this!

---

📁 Part 4: Managing Models & Storage

LM Studio Model Management

• View installed models: My Models tab
• Delete: Right-click → Remove
• Location (Windows): C:\Users\[You]\.lmstudio\models
• Location (Mac): ~/Library/Application Support/lmstudio/models

Ollama Model Management

bash

# List downloaded models
ollama list

# Show model details
ollama show llama3.2

# Delete a model
ollama rm llama3.2

# Model storage location
# Mac/Linux: ~/.ollama/models
# Windows: C:\Users\[You]\.ollama\models

Space-saving tip: Delete models you no longer use. A single 7B model is ~4GB.

---

🧪 Part 5: Testing Your Setup

Quick Benchmark: First Response Time

1. LM Studio: Load model → Ask “Hello” → Measure time to first word
2. Ollama: ollama run llama3.2 "Hello in 3 words"

Typical results (Apple M1/M2):

• 3B models: Instant
• 7B models: 1–3 seconds
• 13B+ models: 5+ seconds

---

🚨 Part 6: Troubleshooting Common Issues

LM Studio Issues

Problem	Solution
Model won’t download	Check internet, firewall, or try different model
“Out of memory”	Close other apps, use smaller model (Phi-3, TinyLlama)
Slow responses	Enable GPU in Settings → Hardware
App crashes	Reinstall, check disk space
No sound/notifications	OS permission settings

Ollama Issues

Problem	Solution
command not found	Reinstall or add to PATH
ollama: command not found (Mac)	brew link ollama
Model download stuck	Ctrl+C → ollama pull [model] again
CUDA errors (Windows)	Install NVIDIA drivers, use WSL2
Metal errors (Mac)	Update macOS, brew upgrade ollama
Port 11434 in use	ollama serve → change port in config

General Tips

• Slow? Use smaller models (3B vs 70B)
• Out of memory? 7B needs ~4GB RAM, 13B needs ~8GB
• GPU not used? Check driver updates
• Confused? Try both tools—they coexist peacefully!

---

🎓 Part 7: What to Do After Setup

Congratulations! You now have a working local LLM. Next steps:

If You Chose LM Studio:

1. [Explore the model library] → Find specialized models for coding, writing, or roleplay
2. [Use local server mode] → Settings → Enable local API
3. [Try our prompt engineering guide] → Get better responses

If You Chose Ollama:

1. [Claude Code setup guide] → The +200% trending workflow
2. [Ollama API tutorial] → Build apps with localhost:11434
3. [Docker deployment] → Containerize your LLM
4. [Open WebUI] → Add a ChatGPT-like interface

---

📋 Quick Reference: Setup Commands & Downloads

Tool	OS	Command / Download	Time
LM Studio	Windows	lmstudio.ai → .exe	2 min
LM Studio	Mac	lmstudio.ai → .dmg	2 min
Ollama	Mac	brew install ollama	30 sec
Ollama	Windows	ollama.com → .exe	1 min
Ollama	Linux	curl -fsSL https://ollama.com/install.sh | sh	1 min
First model	LM Studio	Search + Download	2–10 min
First model	Ollama	ollama run llama3.2	1–5 min

---

❓ Frequently Asked Setup Questions

Q: Do I need an internet connection after setup?
A: No. After models are downloaded, both tools work completely offline.

Q: Which tool uses less RAM/CPU when idle?
A: Ollama. LM Studio keeps some processes running; Ollama unloads models when not in use.

Q: Can I install both?
A: Yes! They don’t conflict. Many users keep both.

Q: How much disk space do I need?
A: Minimum 10GB. Each model is 2GB–8GB. Start with one, delete later.

Q: Why is my first response slow?
A: First load caches the model. Subsequent responses will be faster.

Q: Do I need a powerful computer?
A: 8GB RAM + any modern CPU works for small models (1B–7B). For 13B+, 16GB+ RAM recommended.

Q: Which is better for an old laptop?
A: LM Studio + Phi-3 (3.8B) or TinyLlama (1.1B). Or Ollama + Qwen2.5:0.5B (tiny but fast).

---

🏁 Final Verdict: Which Setup Path Should You Take?

Take the LM Studio path if:

• You want to be chatting in under 5 minutes
• You prefer installers and buttons over commands
• You’re on Windows and want the smoothest experience
• You want to browse and test many models easily

Take the Ollama path if:

• You’re comfortable with a single terminal command
• You use Mac (especially Apple Silicon) or Linux
• You eventually want API access or Docker
• You want Claude Code or Qwen models

Still unsure? Do both. Install LM Studio for quick testing, Ollama for when you’re ready to go deeper. They’re free, and your computer can handle both.