xDev Asia

Build a complete personal AI system with Open WebUI, Continue.dev, Ollama. Integrates Obsidian, VS Code, terminal. Privacy-first workflow.

Running AI Local with Ollama on Apple Silicon

Part 4: Optimization, management & production setup

xdev.asia

Introduction

Final lesson — combine all the knowledge from the previous 11 lessons into a complete workflow. You'll have a privacy-first personal AI system: chat UI, code assistant, note-taking, RAG — all running locally on Mac.

1. Overview architecture

┌─────────────────────────────────────────────┐
│              Personal AI Stack              │
│                                             │
│  ┌──────────┐  ┌───────────┐  ┌──────────┐ │
│  │ Open     │  │ Continue  │  │ Obsidian │ │
│  │ WebUI    │  │ .dev      │  │ + AI     │ │
│  │ (Chat)   │  │ (VS Code) │  │ (Notes)  │ │
│  └────┬─────┘  └─────┬─────┘  └────┬─────┘ │
│       │              │              │       │
│       └──────────────┼──────────────┘       │
│                      │                      │
│              ┌───────┴──────┐               │
│              │   Ollama     │               │
│              │ localhost:   │               │
│              │   11434      │               │
│              └───────┬──────┘               │
│                      │                      │
│              ┌───────┴──────┐               │
│              │ Apple Silicon│               │
│              │ GPU + Neural │               │
│              │   Engine     │               │
│              └──────────────┘               │
└─────────────────────────────────────────────┘

2. Open WebUI — ChatGPT-like UI for Ollama

Open WebUI is the most beautiful web UI for Ollama, running locally.

Install

# Cách 1: Docker (khuyến nghị)
docker run -d \
  --name open-webui \
  -p 3000:8080 \
  -v open-webui:/app/backend/data \
  --add-host=host.docker.internal:host-gateway \
  -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \
  --restart always \
  ghcr.io/open-webui/open-webui:main

# Cách 2: pip (không cần Docker)
pip3 install open-webui
open-webui serve --port 3000

Access: http://localhost:3000

Main features

Multi-model chat: Switch between models
Chat history: Save and search conversations
Markdown: Render code blocks, tables, math
File upload: RAG with documents
Image generation: Integrates Stable Diffusion
Voice: Speech-to-text and text-to-speech
Preset prompts: Library system prompts
Multi-user: Supports multiple users
Mobile-friendly: Responsive UI

Good configuration

Model presets: Settings → Models → create presets for each use case
System prompts: Settings → General → Default system prompt
RAG: Upload PDF/TXT → AI answers based on the document
Keyboard shortcuts: Ctrl+Shift+O → New chat

3. Continue.dev — AI assistant in VS Code

Continue is the most powerful AI extension for VS Code, supporting Ollama.

Install

VS Code → Extensions → find "Continue"
Install Continue - Codestral, Claude, and more

Configuration for Ollama

Create/edit ~/.continue/config.yaml:

models:
  - name: Ollama - Llama 3.2
    provider: ollama
    model: llama3.2
    apiBase: http://localhost:11434

  - name: Ollama - Qwen Coder
    provider: ollama
    model: qwen2.5-coder:7b
    apiBase: http://localhost:11434

  - name: Ollama - Gemma 3
    provider: ollama
    model: gemma3:12b
    apiBase: http://localhost:11434

tabAutocompleteModel:
  provider: ollama
  model: qwen2.5-coder:1.5b
  apiBase: http://localhost:11434

embeddingsProvider:
  provider: ollama
  model: nomic-embed-text
  apiBase: http://localhost:11434

Use

Features	Shortcuts	Description
Chat	`Cmd+L`	Open chat panel
Inline edit	`Cmd+I`	Edit code inline
Select + explain	Select → `Cmd+L`	Code explanation
Autocomplete tab	`Tab`	Code suggestions
Terminal debug	Click error	AI explains errors

Custom slash commands

# ~/.continue/config.yaml
customCommands:
  - name: review
    description: Review selected code
    prompt: |
      Review code này. Tìm bugs, security issues, performance problems.
      Suggest improvements. Trả lời bằng tiếng Việt.

  - name: test
    description: Generate tests
    prompt: |
      Viết unit tests cho code này dùng pytest.
      Cover edge cases. Mock external dependencies.

  - name: doc
    description: Add documentation
    prompt: |
      Thêm docstring và comments cho code này.
      Dùng Google style docstring.

4. Terminal workflow

Useful Alias

Added ~/.zshrc:

# Quick chat
alias ai="ollama run llama3.2"
alias aicode="ollama run qwen2.5-coder:7b"
alias aiviet="ollama run gemma3:4b 'Trả lời bằng tiếng Việt:'"

# Pipe output to AI
alias explain="ollama run llama3.2 'Giải thích output này:'"
alias fix="ollama run qwen2.5-coder:7b 'Fix lỗi này:'"

# Quick functions
function ai-commit() {
    git diff --staged | ollama run llama3.2 "Write a concise git commit message for this diff. Follow conventional commits format. Just the message, nothing else."
}

function ai-review() {
    cat "$1" | ollama run qwen2.5-coder:7b "Review code này, tìm bugs và đề xuất cải thiện. Tiếng Việt."
}

function ai-explain() {
    cat "$1" | ollama run llama3.2 "Giải thích file code này đang làm gì. Tiếng Việt, ngắn gọn."
}

Use

# Quick chat
ai "Docker volume mount là gì?"

# Pipe command output
kubectl get pods | explain

# Generate commit message
git add . && ai-commit

# Review file
ai-review src/app.py

# Explain file
ai-explain config/nginx.conf

5. Obsidian + AI Notes

Obsidian + AI plugin creates "second brain" with local AI.

Plugin: Smart Connections

Obsidian → Settings → Community plugins → find "Smart Connections"
Install and enable
Settings → Smart Connections:
- Chat Model Platform: Ollama
- Chat Model: llama3.2
- Embedding Model Platform: Ollama
- Embedding Model: nomic-embed-text

Plugin: Copilot for Obsidian

Install "Copilot" plugin
Settings → Default Model Provider: Ollama
Ollama URL: http://localhost:11434

Workflow

Đọc tài liệu → Highlight → AI tóm tắt → Lưu note → AI search notes
     ↓              ↓            ↓             ↓           ↓
   Input        Selection    Ollama API    Obsidian    Embeddings
                                                      + RAG local

6. RAG with personal documents

Ollama + ChromaDB

pip3 install chromadb ollama

#!/usr/bin/env python3
"""Personal RAG - Index và search tài liệu local."""

import ollama
import chromadb
from pathlib import Path

# Setup ChromaDB
client = chromadb.PersistentClient(path="./chroma_db")
collection = client.get_or_create_collection("my_docs")

def index_folder(folder_path):
    """Index tất cả .md và .txt files."""
    folder = Path(folder_path)
    files = list(folder.glob("**/*.md")) + list(folder.glob("**/*.txt"))

    for f in files:
        content = f.read_text(encoding='utf-8')

        # Chunk by paragraphs
        chunks = [c.strip() for c in content.split('\n\n') if c.strip()]

        for i, chunk in enumerate(chunks):
            doc_id = f"{f.stem}_{i}"

            # Generate embedding
            response = ollama.embeddings(
                model='nomic-embed-text',
                prompt=chunk
            )

            collection.upsert(
                ids=[doc_id],
                embeddings=[response['embedding']],
                documents=[chunk],
                metadatas=[{"source": str(f), "chunk": i}]
            )

    print(f"✅ Indexed {len(files)} files")

def search(query, top_k=3):
    """Search documents."""
    query_emb = ollama.embeddings(
        model='nomic-embed-text',
        prompt=query
    )['embedding']

    results = collection.query(
        query_embeddings=[query_emb],
        n_results=top_k
    )

    return results

def ask(question):
    """RAG: search + generate."""
    results = search(question)
    context = "\n\n".join(results['documents'][0])

    response = ollama.chat(
        model='llama3.2',
        messages=[{
            'role': 'user',
            'content': f"""Dựa vào context sau để trả lời câu hỏi.
Nếu không tìm thấy thông tin, nói rõ.

Context:
{context}

Câu hỏi: {question}

Trả lời bằng tiếng Việt:"""
        }]
    )
    return response['message']['content']

# Sử dụng
if __name__ == "__main__":
    # Index notes
    index_folder("~/Documents/notes")

    # Ask questions
    answer = ask("Kubernetes pod restart policy có mấy loại?")
    print(answer)

7. Automation scripts

Daily AI summary

#!/usr/bin/env python3
"""Tóm tắt git commits hàng ngày."""

import subprocess
import ollama
from datetime import datetime

def get_today_commits():
    result = subprocess.run(
        ['git', 'log', '--since=midnight', '--pretty=format:%s'],
        capture_output=True, text=True
    )
    return result.stdout

def daily_summary():
    commits = get_today_commits()
    if not commits:
        print("Không có commits hôm nay.")
        return

    response = ollama.chat(
        model='llama3.2',
        messages=[{
            'role': 'user',
            'content': f"""Tóm tắt những thay đổi trong ngày dựa trên git commits:

{commits}

Viết tóm tắt ngắn gọn bằng tiếng Việt, gom nhóm theo feature/fix."""
        }]
    )

    today = datetime.now().strftime('%Y-%m-%d')
    print(f"📋 Daily Summary - {today}")
    print("=" * 40)
    print(response['message']['content'])

daily_summary()

Auto README generator

#!/usr/bin/env python3
"""Generate README.md từ cấu trúc project."""

import ollama
import subprocess

def get_project_info():
    # File structure
    tree = subprocess.run(
        ['find', '.', '-maxdepth', '3', '-not', '-path', './.git/*',
         '-not', '-path', './node_modules/*', '-not', '-path', './.next/*'],
        capture_output=True, text=True
    ).stdout

    # Package info
    import json
    try:
        with open('package.json') as f:
            pkg = json.load(f)
    except FileNotFoundError:
        pkg = {}

    return tree, pkg

def generate_readme():
    tree, pkg = get_project_info()

    response = ollama.chat(
        model='llama3.2',
        messages=[{
            'role': 'user',
            'content': f"""Generate README.md cho project này.

File structure:
{tree[:2000]}

Package.json:
{str(pkg)[:1000]}

Include sections:
- Project title & description
- Features
- Tech stack
- Getting started (install, run)
- Project structure
- Contributing
- License

Viết bằng tiếng Anh, format markdown."""
        }]
    )

    with open('README.ai.md', 'w') as f:
        f.write(response['message']['content'])
    print("✅ Generated README.ai.md")

generate_readme()

8. Security & Privacy checklist

What HAS BEEN achieved

✅ Everything runs locally — no data is sent to the internet ✅ Does not depend on the cloud — works offline ✅ No tracking — no telemetry from model ✅ You own the data — the models and data are on your device ✅ No rate limit — use as much as you like ✅ No cost — free, forever

Best practices

# 1. Ollama chỉ bind localhost (default, an toàn)
export OLLAMA_HOST=127.0.0.1:11434

# 2. Nếu cần expose ra mạng, dùng authentication
# (Open WebUI có built-in auth)

# 3. Encrypt sensitive data
# macOS FileVault đã mã hóa toàn bộ disk

# 4. Regular backup models list
ollama list > ~/backups/ollama-models-$(date +%Y%m%d).txt

Compare with cloud AI

	Local (Ollama)	Cloud (OpenAI, etc.)
Cost	$0	$20-200/month
Privacy	100% local	Data sent to the internet
Speed	Hardware dependent	Fast, stable
Quality	7B ≈ GPT-3.5	GPT-4, Claude 3.5
Offline	✅	❌
Rate limit	No	Yes
Custom model	✅ Modelfile	❌ Limited

9. Complete setup script

#!/bin/bash
# setup-ai-workspace.sh
# One-click Personal AI Setup trên macOS

set -e

echo "🚀 Setting up Personal AI Workspace..."

# 1. Install Ollama
if ! command -v ollama &> /dev/null; then
    echo "📦 Installing Ollama..."
    curl -fsSL https://ollama.com/install.sh | sh
fi

# 2. Pull recommended models
echo "📥 Pulling models..."
ollama pull llama3.2
ollama pull qwen2.5-coder:7b
ollama pull nomic-embed-text
ollama pull gemma3:4b

# 3. Set environment variables
echo "⚙️  Configuring environment..."
cat >> ~/.zshrc << 'ZSHRC'

# === Ollama AI Config ===
export OLLAMA_NUM_PARALLEL=2
export OLLAMA_FLASH_ATTENTION=1
export OLLAMA_KEEP_ALIVE=10m

# AI aliases
alias ai="ollama run llama3.2"
alias aicode="ollama run qwen2.5-coder:7b"
function ai-commit() {
    git diff --staged | ollama run llama3.2 "Write a concise conventional commit message for this diff. Just the message."
}
ZSHRC

# 4. Create custom models
echo "🔧 Creating custom models..."

cat > /tmp/Modelfile.xdev << 'EOF'
FROM llama3.2
SYSTEM """Bạn là xDev AI, trợ lý lập trình chuyên nghiệp.
- Trả lời bằng tiếng Việt
- Code rõ ràng, có comment
- Đề cập best practices"""
PARAMETER temperature 0.5
PARAMETER num_ctx 4096
EOF

ollama create xdev-ai -f /tmp/Modelfile.xdev

# 5. Start Open WebUI (nếu có Docker)
if command -v docker &> /dev/null; then
    echo "🌐 Starting Open WebUI..."
    docker run -d \
        --name open-webui \
        -p 3000:8080 \
        -v open-webui:/app/backend/data \
        --add-host=host.docker.internal:host-gateway \
        -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \
        --restart always \
        ghcr.io/open-webui/open-webui:main 2>/dev/null || true
fi

# 6. Install Python packages
echo "🐍 Installing Python packages..."
pip3 install ollama chromadb rich 2>/dev/null || true

echo ""
echo "✅ Setup complete!"
echo ""
echo "📋 What's ready:"
echo "   • Ollama: http://localhost:11434"
echo "   • Open WebUI: http://localhost:3000 (if Docker)"
echo "   • Models: llama3.2, qwen2.5-coder:7b, gemma3:4b"
echo "   • Custom: xdev-ai"
echo "   • Aliases: ai, aicode, ai-commit"
echo ""
echo "🎯 Quick start:"
echo "   ai 'Hello, AI!'"
echo "   aicode 'Viết function sort bằng Python'"
echo ""
echo "🎉 Happy coding with local AI!"

10. Course summary

What did you learn?

Article	Content	Skills
1	Apple Silicon & AI	Understanding hardware
2	Install Ollama	Setup & CLI
3	Select model	Model selection
4	MLX Framework	MLX basics
5	mlx-lm	Quantization, conversion
6	Ollama + MLX	Integration
7	REST API	HTTP API, OpenAI-compatible
8	Python integration	Chatbot, RAG
9	Vision models	Image analysis
10	Optimize performance	RAM, concurrency
11	Modelfiles	Custom models
12	Complete Workflow	Production setup

Follow xDev.asia to update new series
Join community to exchange experiences
Contribute — share your Modelfile, script, workflow

Final exercises

Full setup: Ollama + Open WebUI + Continue.dev
Create 3 custom models for individual workflows
Build a RAG system index of your documents (notes, docs)
Write an ai-commit script and use it every day for 1 week
(Bonus) Create a Telegram/Discord bot running locally using Ollama

🎉 Congratulations on completing the course!

You now have the knowledge to run AI completely locally on Apple Silicon — fast, free, and completely secure.

Lesson 12: Complete Workflow - Personal AI setup for developers

Introduction

1. Overview architecture

2. Open WebUI — ChatGPT-like UI for Ollama

Install

Main features

Good configuration

3. Continue.dev — AI assistant in VS Code

Install

Configuration for Ollama

Use

Custom slash commands

4. Terminal workflow

Useful Alias

Use

5. Obsidian + AI Notes

Plugin: Smart Connections

Plugin: Copilot for Obsidian

Workflow

6. RAG with personal documents

Ollama + ChromaDB

7. Automation scripts

Daily AI summary

Auto README generator

8. Security & Privacy checklist

What HAS BEEN achieved

Best practices

Compare with cloud AI

9. Complete setup script

10. Course summary

What did you learn?

Next

Final exercises