Skip to main content

​
Quickstart Guide

Get started with Ollama and run your first large language model locally in just a few minutes.
This guide assumes you have already installed Ollama. If not, see the Installation Guide first.

​
Your First Model

1

Run Ollama

The simplest way to get started is to run a model directly:
ollama run gemma3
This command will:
  • Download the Gemma 3 model (if not already downloaded)
  • Start the Ollama server (if not already running)
  • Begin an interactive chat session
On first run, the model will be downloaded. This may take a few minutes depending on your internet connection.
2

Chat with the Model

Once the model is loaded, you’ll see a prompt:
>>> Send a message (/? for help)
Try asking a question:
>>> Why is the sky blue?
The model will stream its response in real-time.
Type /help to see all available chat commands, or press Ctrl+D to exit.
3

Try Other Models

Explore different models from the library:
# Run Llama 3.2 (Meta's model)
ollama run llama3.2

# Run Mistral (efficient and fast)
ollama run mistral

# Run Phi-3 (Microsoft's compact model)
ollama run phi3
Browse all available models →

​
Using the REST API

Ollama provides a REST API server that runs on http://localhost:11434 by default.

​
Generate a Response

curl http://localhost:11434/api/generate -d '{
  "model": "gemma3",
  "prompt": "Why is the sky blue?"
}'

​
Chat with Context

Maintain conversation history: 
curl http://localhost:11434/api/chat -d '{
  "model": "gemma3",
  "messages": [
    {
      "role": "user",
      "content": "Why is the sky blue?"
    }
  ],
  "stream": false
}'

​
Essential Commands

​
List Downloaded Models

ollama list
Output: 
NAME              ID            SIZE      MODIFIED
gemma3:latest     a80c4f17acd5  2.0 GB    2 hours ago
llama3.2:latest   0a8c26691023  4.7 GB    1 day ago

​
Pull a Model

Download a model without running it: 
ollama pull mistral

​
Remove a Model

Free up disk space: 
ollama rm mistral

​
Check Running Models

See which models are currently loaded in memory: 
ollama ps
Output: 
NAME              ID            SIZE      PROCESSOR    CONTEXT    UNTIL
gemma3:latest     a80c4f17acd5  2.0 GB    100% GPU     4096       4 minutes from now

​
Multimodal Models

Some models can process images along with text: 
ollama run llava "What's in this image? /path/to/image.jpg"
Or via the API: 
curl http://localhost:11434/api/generate -d '{
  "model": "llava",
  "prompt": "What is in this image?",
  "images": ["base64_encoded_image_data"]
}'

​
Non-Interactive Mode

Run one-off prompts without entering chat mode: 
# Single prompt
ollama run gemma3 "Explain quantum computing in simple terms"

# Pipe input
echo "Write a haiku about programming" | ollama run gemma3

# Save output
ollama run gemma3 "Generate a Python function" > output.py

​
Model Options

Customize model behavior with options: 
curl http://localhost:11434/api/generate -d '{
  "model": "gemma3",
  "prompt": "Tell me a story",
  "options": {
    "temperature": 0.8,
    "top_p": 0.9,
    "seed": 42
  }
}'
  • temperature (0.0-2.0): Controls randomness (default: 0.8)
  • top_p (0.0-1.0): Nucleus sampling threshold (default: 0.9)
  • seed: Set for reproducible outputs
  • num_predict: Maximum tokens to generate
  • stop: Custom stop sequences

​
Keep Alive

Control how long models stay in memory: 
# Keep loaded for 10 minutes
ollama run gemma3 --keepalive 10m

# Unload immediately after use
ollama run gemma3 --keepalive 0

# Keep loaded indefinitely
ollama run gemma3 --keepalive -1
Default is 5 minutes after last use.

​
Server Management

​
Start the Server

The server starts automatically when you run a model, but you can start it manually: 
ollama serve
The server runs on http://localhost:11434 by default.

​
Environment Variables

  • OLLAMA_HOST: Change the bind address (default: 127.0.0.1:11434)
  • OLLAMA_MODELS: Custom model storage location
  • OLLAMA_NUM_PARALLEL: Number of parallel requests (default: 1)
  • OLLAMA_MAX_LOADED_MODELS: Max models in memory (default: 1)
  • OLLAMA_DEBUG: Enable debug logging
Example: 
export OLLAMA_HOST=0.0.0.0:11434
ollama serve

​
Troubleshooting

Model sizes range from 2GB to 70GB+. Use a wired connection for faster downloads. You can resume interrupted downloads by running the same command again.
Try a smaller quantized model:
  • Use gemma3:2b instead of gemma3:8b
  • Use Q4 quantization: llama3.2:3b-q4_0
  • Reduce context size with num_ctx option
The server may not be running:
ollama serve
Or check if another process is using port 11434.
On Linux, ensure NVIDIA drivers are installed. On macOS, Metal should work automatically. Check with:
ollama run gemma3 --verbose

​
What’s Next?

Create Custom Models

Customize models with system prompts and parameters

API Reference

Complete REST API documentation

Import Models

Import models from PyTorch or Safetensors

CLI Reference

Complete command-line documentation
Join the Discord community for help and to share your projects!

Build docs developers (and LLMs) love

Get started for freeTalk to us