Build Your Second Brain with Free Local AI (Obsidian + Ollama Guide)

The promise of AI-powered note-taking is compelling: ask questions about your knowledge base, get intelligent summaries, and have a copilot that understands your personal context. The problem? Most solutions require sending your private notes to external APIs.

This guide shows you how to set up a 100% local AI inside Obsidian using Ollama. Everything runs on your Mac — no internet required after setup, no subscription fees, complete privacy.

What You’ll Build

By the end of this guide, you’ll have:

• AI-powered chat within Obsidian (ask questions, get summaries)
• Semantic vault search using RAG (Retrieval-Augmented Generation)
• Complete privacy — all processing happens locally

Architecture Overview

📝 Obsidian + Copilot Plugin
          ↕️
      HTTP Request
          ↕️
🚀 Ollama Server (localhost:11434)
          │
    ┌─────┴─────┐
    ↓           ↓
🦙 llama3.2    🔍 nomic-embed-text
   (Chat)         (Embeddings)

Prerequisites

• macOS (Apple Silicon or Intel)
• Obsidian installed
• ~3GB disk space for models
• 8GB+ RAM recommended

---

Step 1: Install Ollama

Download and install from ollama.ai. After installation, you’ll see the Ollama icon in your menu bar.

Step 2: Download the Models

Open Terminal and run:

# Chat model (~2GB)
ollama pull llama3.2

# Embedding model for search (~274MB)
ollama pull nomic-embed-text

Verify with:

ollama list

Step 3: Allow Ollama to Talk to Obsidian

This step is critical — without it, Obsidian can’t connect to Ollama:

launchctl setenv OLLAMA_ORIGINS "app://obsidian.md*"

Then quit and restart Ollama from the menu bar.

Step 4: Install the Copilot Plugin

1. Open Obsidian → Settings → Community Plugins → Browse
2. Search for “Copilot”
3. Install and enable “Copilot for Obsidian”

Step 5: Configure Copilot

Add the Chat Model

1. Settings → Copilot → Custom Models → Add Custom Model
2. Configure:
   • Model name: llama3.2
   • Provider: Ollama
   • Base URL: http://localhost:11434
   • API Key: Leave blank

Add the Embedding Model

1. Add another custom model:
   • Model name: nomic-embed-text
   • Provider: Ollama

Set Defaults

In Copilot’s Basic tab:

• Chat Model: llama3.2
• Embedding Model: nomic-embed-text

Step 6: Index Your Vault

1. Press Cmd+P to open Command Palette
2. Run: Copilot: Index Vault
3. Wait for indexing to complete

---

Testing Your Setup

Verify Ollama is running:

curl http://localhost:11434
# Should return: Ollama is running

Start the chat model:

ollama run llama3.2

Verify models are loaded:

ollama ps

Now open the Copilot sidebar in Obsidian and ask a question about your notes!

---

Troubleshooting

”Connection Refused” Error

Re-run the launchctl setenv command and restart Ollama.

”Embedding EOF” Error

Make sure models are running with ollama run llama3.2.

General Debugging

Open Obsidian’s console with Cmd+Option+I and check for errors.

---

Alternative Models

# Smaller/faster (1B parameters)
ollama pull llama3.2:1b

# Larger/smarter (requires more RAM)
ollama pull llama3.1:8b

# Code-focused
ollama pull codellama

---

Why This Matters

Benefit	Details
Cost	Free forever — no API fees
Privacy	Notes never leave your Mac
Offline	Works without internet
Performance	Optimized for Apple Silicon

---

Quick Reference Checklist

• Ollama installed and running
• llama3.2 and nomic-embed-text pulled
• OLLAMA_ORIGINS set and Ollama restarted
• Copilot plugin installed
• Models configured in Copilot settings
• Vault indexed
• Chat working

---

Your second brain is now AI-powered — and it’s entirely yours.