Quartz 4

Ollama Setup Guide

3 min read

🤖 Ollama Local AI Setup

This guide will help you set up Ollama to power the weekly analysis feature.

What is Ollama?

Ollama allows you to run large language models locally on your machine. This means:

  • âś… Complete privacy - your data never leaves your computer
  • âś… No API costs
  • âś… Works offline
  • âś… Fast responses

Installation

macOS

  1. Download Ollama:

    • Visit https://ollama.ai
    • Click “Download for macOS”
    • Open the downloaded .dmg file and drag Ollama to Applications

  2. Verify Installation:

    ollama --version

  3. Pull a Model:

    # Recommended: Llama 3.2 (1B - fast, good for analysis)
ollama pull llama3.2
 
# Or for better quality (larger model):
ollama pull llama3.2:3b

Alternative Models

You can use different models depending on your hardware:

# Lightweight (fast, uses ~1GB RAM)
ollama pull llama3.2:1b
 
# Medium (balanced, ~3GB RAM)
ollama pull llama3.2:3b
 
# Large (best quality, ~8GB RAM)
ollama pull llama3.1:8b
 
# Code-focused (for technical analysis)
ollama pull codellama

Configuration

Update the Script

If you want to use a different model, edit /content/_System/Scripts/obsidian/analyze-weekly.js:

Find this line:

model: 'llama3.2', // Change to your preferred model

Change it to your preferred model:

model: 'llama3.2:3b', // or 'codellama', 'llama3.1:8b', etc.

Usage

Running Ollama

Ollama runs automatically in the background after installation. You can verify it’s running:

# Check if Ollama is running
curl http://localhost:11434/api/tags
 
# Or use the Ollama command
ollama list

Using Weekly Analysis

  1. Open your Dashboard
  2. Click the ”🧠 Analyze Week (AI)” button
  3. Wait for the analysis to complete
  4. A new report will be created with AI-powered insights

Troubleshooting

”Ollama not available” Error

Issue: The script shows a fallback message saying Ollama isn’t available.

Solutions:

  1. Check if Ollama is running:

    ollama list

  2. Restart Ollama:

    • macOS: Open Applications → Ollama
    • Or run: ollama serve

  3. Check the port (should be 11434):

    lsof -i :11434

Model Not Found

Issue: Error says model doesn’t exist.

Solution:

# List available models
ollama list
 
# Pull the model if missing
ollama pull llama3.2

Performance Issues

Issue: Analysis is slow or hangs.

Solutions:

  1. Use a smaller model:

    ollama pull llama3.2:1b

  2. Reduce context in the script (edit num_predict in analyze-weekly.js):

    options: {
    temperature: 0.7,
    num_predict: 1000  // Reduced from 2000
}

  3. Close other applications to free up memory

Port Conflict

Issue: Port 11434 is already in use.

Solutions:

  1. Stop other processes using the port:

    lsof -ti:11434 | xargs kill -9

  2. Or change Ollama’s port (advanced):

    OLLAMA_HOST=0.0.0.0:11435 ollama serve

    Then update the script URL to http://localhost:11435/api/generate

Advanced Configuration

Temperature Setting

Controls creativity vs consistency:

temperature: 0.3  // More focused, consistent
temperature: 0.7  // Balanced (default)
temperature: 1.0  // More creative, varied

Context Length

How much text to process:

num_predict: 500   // Short, quick analysis
num_predict: 2000  // Detailed analysis (default)
num_predict: 4000  // Very detailed (slower)

Custom Prompts

Edit the analysisPrompt in analyze-weekly.js to customize what the AI focuses on:

const analysisPrompt = `You are analyzing a weekly review...
 
Focus on:
1. Technical tasks vs creative ideas
2. Blocked tasks requiring help
3. Quick wins that can be completed today
 
...`;

Resources

Testing Your Setup

Run this command to test Ollama:

ollama run llama3.2 "Summarize: I need to build a component for user authentication, create API endpoints, and fix the dashboard layout"

You should get a concise summary back. If this works, the weekly analysis button will work!

Quick Reference

# Essential Commands
ollama list                    # Show installed models
ollama pull MODEL_NAME         # Download a model
ollama rm MODEL_NAME           # Remove a model
ollama serve                   # Start Ollama server
ollama ps                      # Show running models

For issues, check the Ollama documentation or the script’s console output in Obsidian DevTools (Cmd+Option+I)

Graph View

Backlinks