MacAI > AI Knowledge Hub
AI Knowledge Hub
AI Knowledge Hub
Understand the tools.
Master your AI stack.
Every MacAI workstation is built on proven open-source AI technologies. Here is a technical guide to the key concepts, tools, and frameworks powering your setup.
OpenClaw — Personal AI Agent
OpenClaw is an open-source autonomous AI agent that runs locally on your Mac and connects to your messaging apps (WhatsApp, Telegram, Discord, Slack, Signal, iMessage). Unlike a chatbot, it can execute real tasks — manage files, browse the web, automate workflows, and control apps on your behalf.
Installation on Mac
# Install Node.js 22+ via Homebrew
brew install node@22
# Install OpenClaw
curl -fsSL https://openclaw.ai/install.sh | bash
# Run onboarding wizard
openclaw onboard --install-daemon
# Configure AI provider (Claude recommended)
openclaw config set provider anthropic
openclaw config set api-key YOUR_KEY
Key Capabilities
- ✓ Multi-channel messaging (WhatsApp, Telegram, Discord, Signal)
- ✓ Local-first — memory stored as Markdown on your disk
- ✓ 15+ built-in AgentSkills (shell, files, web, calendar)
- ✓ Browser automation and web interaction
- ✓ Voice wake and talk mode
- ✓ ClawHub skill marketplace — hundreds of community extensions
- ✓ MCP server integration for external tool access
MCP — Model Context Protocol
MCP is an open standard developed by Anthropic for connecting AI models to external tools and data sources. Think of it as USB-C for AI — a universal plug that lets any AI model talk to any service (Google Calendar, Notion, databases, APIs, file systems) through a standardized protocol.
How It Works
An MCP server exposes tools (functions) that AI models can call. Your AI asks "what tools are available?", discovers them, and uses them to complete tasks — reading files, querying databases, or triggering actions.
Use Cases
Connect AI to Google Workspace, Slack, Notion, Home Assistant, GitHub, databases, or any custom API. The AI can read your calendar, search your docs, post messages, or control smart devices.
Local MCP on Your Mac
We configure local MCP servers on your workstation so your AI agent can access your files, databases, and services — all without sending data to the cloud.
RAG — Retrieval-Augmented Generation
RAG lets your AI search through your own documents, databases, and knowledge bases before answering. Instead of relying only on what the model was trained on, RAG retrieves relevant context from your private data — making responses accurate, current, and specific to your business.
How RAG Works
1. Ingest — Your documents (PDFs, emails, contracts, manuals) are split into chunks
2. Embed — Each chunk is converted to a vector (numerical representation) using Nomic Embed
3. Store — Vectors are saved in a local vector database (ChromaDB or PostgreSQL+pgvector)
4. Query — When you ask a question, relevant chunks are retrieved and fed to the LLM as context
5. Generate — The LLM answers using your specific data — accurate, cited, and private
RAG Tools We Install
- PrivateGPT — Drop-in document Q&A, 100% local
- Nomic Embed — Text embedding model for vector search
- ChromaDB — Lightweight vector database
- pgvector — PostgreSQL extension for enterprise vector search
- Open WebUI RAG — Upload documents directly into chat for instant Q&A
AI Skills — Extend Your Agent
AI Skills are modular capabilities you add to your AI agent. Each skill is defined in a SKILL.md file — a portable, human-readable instruction set that teaches the agent how to perform a specific task. Skills can be installed from community marketplaces, shared between agents, or custom-built for your business.
Core Skill Categories
Shell & System
Execute commands, manage processes, system monitoring
Web & Browser
Browse pages, fill forms, scrape data, interact with sites
Files & Documents
Read, write, organize, convert, and search documents
APIs & Integration
Connect to external services, trigger webhooks, sync data
How AI Skills Work
1. Each skill lives in a folder with a SKILL.md file — the instruction set your AI reads
2. When triggered, the agent loads the skill's instructions and follows the defined workflow
3. Skills can call tools (shell, browser, APIs), chain together, and return structured output
4. Skills are versioned, shareable, and can be published to ClawHub or kept private
AI skills turn your agent from a chatbot into a worker. Instead of asking "how do I convert this PDF?", your agent just does it — using the skill's instructions to pick the right tool, run the right commands, and deliver the result.
Example: SKILL.md File
# skills/pdf-converter/SKILL.md
name: PDF Converter
trigger: "convert * to PDF"
tools: [shell, files]
## Instructions
When the user asks to convert a document
to PDF, follow these steps:
1. Identify the input file type
2. Use the appropriate converter:
- .docx → pandoc or libreoffice
- .html → wkhtmltopdf
- .md → pandoc with LaTeX
3. Apply formatting preferences
4. Save to the user's output folder
5. Confirm with file size & page count
## Error Handling
If the tool is missing, install it via
Homebrew and retry once.
Advanced AI Skills
Reasoning & Analysis
Multi-step research, data analysis, report generation, competitive intelligence. The agent breaks complex problems into sub-tasks and synthesizes findings.
Creative & Content
Generate presentations, draft proposals, write marketing copy, create social media content, produce newsletters — all following your brand voice and style guidelines.
Data & Automation
Parse spreadsheets, clean datasets, generate charts, ETL pipelines, scheduled reports, database queries. Turn raw data into actionable insights automatically.
Communication
Draft emails, schedule meetings, summarize threads, translate messages, manage contacts. Your AI handles the back-and-forth so you can focus on decisions.
Security & Privacy
PII detection, document redaction, access control checks, compliance scanning, encrypted storage management. Keep your data safe while AI works on it.
DevOps & Coding
Code review, Git operations, CI/CD management, Docker orchestration, infrastructure monitoring, automated testing. Your AI becomes a junior DevOps engineer.
Build Your Own AI Skills
Every MacAI workstation includes a skill authoring environment. Create custom SKILL.md files tailored to your exact workflows — from invoice processing to client onboarding. We help you design, test, and deploy production-ready skills during setup. Enterprise tiers include up to 10 custom skills built by our team.
Plugins & Extensions
Plugins are code modules that extend your AI agent with new commands, tools, and integrations. They go beyond simple skills by adding persistent services, custom UI, and deep integrations with third-party platforms.
OpenClaw ClawHub
Community skill marketplace — browse, install, and share skills. Your agent can auto-discover and install skills from ClawHub on demand.
n8n Workflows
Visual automation builder with 400+ integrations — connect AI to email, CRM, databases, and business tools with drag-and-drop workflow design.
Custom Development
We build bespoke plugins and integrations tailored to your business — connecting your AI to proprietary systems, internal APIs, and custom workflows.
LLM Models & Ollama
Large Language Models (LLMs) are the core intelligence behind your AI workstation. Ollama is the runtime engine that lets you download and run these models locally on your Mac — no cloud required, no API fees, full privacy. Your Mac Studio's unified memory lets you run models that would require expensive GPU clusters elsewhere.
Pre-Installed Models
- Llama 3.1 405B — Meta's flagship, near GPT-4 performance, runs locally on 512GB Mac Studio
- DeepSeek V3 — Mixture-of-experts model, exceptional at coding and reasoning tasks
- Mistral Large — Multilingual powerhouse with 128K context window, strong for enterprise
- Qwen 2.5 72B — Alibaba's top model, excellent Chinese & English bilingual capability
- Llama 3.2 Vision — Multimodal — understands images, screenshots, documents, and charts
- Nomic Embed — Specialized embedding model for RAG vector search
Quantization & Performance
Quantization compresses models to use less memory while retaining most of their intelligence. This is how your Mac Studio can run 405B parameter models that would normally need a server farm.
- ✓ Q4_K_M — 4-bit quantization, best memory/quality balance
- ✓ Q8_0 — 8-bit, highest quality for critical tasks
- ✓ FP16 — Full precision for fine-tuning and research
- ✓ Hot-swap between models in seconds via Ollama CLI
Prompt Engineering
Prompt engineering is the art of writing instructions that get the best results from AI models. A well-crafted prompt can be the difference between a generic response and expert-level output. We configure optimized system prompts for every tool on your workstation.
System Prompts
Persistent instructions that define your AI's personality, role, and rules. Every tool gets a custom system prompt — your chat assistant knows your industry, your code reviewer knows your standards.
Core Techniques
Chain-of-thought reasoning, few-shot examples, XML-tagged structure, role assignment, temperature control, and output format specification. These techniques dramatically improve accuracy and consistency.
Prompt Library
Every MacAI workstation ships with a curated prompt library — tested templates for summarization, translation, code review, data extraction, creative writing, and business analysis, ready to use or customize.
AI Agents & Agentic Workflows
AI agents go beyond simple chat — they can plan, decide, use tools, and execute multi-step tasks autonomously. An agent receives a goal, breaks it into steps, picks the right tools, handles errors, and delivers results — like a digital employee that works 24/7.
The Agent Loop
1. Observe — Receive a goal or trigger from the user or a schedule
2. Plan — Break the task into logical sub-steps and identify required tools
3. Act — Execute each step using skills, MCP tools, APIs, or shell commands
4. Reflect — Evaluate results, handle errors, and adjust the plan if needed
5. Deliver — Return the final output and report what was done
What Agents Can Do
- ✓ Research a topic, compile findings, and write a report
- ✓ Monitor inbox, categorize emails, and draft replies
- ✓ Scrape competitor pricing and update your spreadsheet
- ✓ Process invoices, extract data, and update your accounting system
- ✓ Multi-agent workflows — agents coordinating to complete complex projects
Local AI on Apple Silicon
Apple Silicon's unified memory architecture is a game-changer for local AI. Unlike traditional GPUs where VRAM is separate and limited, the M5 Ultra shares its entire 512GB memory pool between CPU and GPU — meaning you can load massive models that would cost thousands per month on cloud GPU instances.
MLX Framework
Apple's open-source ML framework built specifically for Apple Silicon. Run and fine-tune models with native Metal GPU acceleration — training custom models on your own data, right on your desk.
Unified Memory Advantage
512GB unified memory means no VRAM bottleneck. Load 405B parameter models that need 240GB+ memory. Run multiple models simultaneously. No cloud bills, no latency, no data leaving your machine.
Performance Benchmarks
M5 Ultra delivers 30+ tokens/second on 70B models and 10+ tok/s on 405B. Fine-tune LoRA adapters in minutes, not hours. Run Stable Diffusion XL at 2-3 seconds per image. Silent, efficient, always on.
Image Generation
Generate professional images, illustrations, product mockups, and marketing visuals — entirely on your Mac, without cloud subscriptions. Stable Diffusion and FLUX run natively on Apple Silicon with Metal acceleration for fast, private image creation.
Stable Diffusion XL
Industry-standard open-source image model. Generate photorealistic images, illustrations, logos, and concept art from text prompts. Fine-tune with your own brand assets using LoRA adapters.
ComfyUI Workflow
Visual node-based editor for building complex image generation pipelines. Chain models, apply controlnets, run img2img, upscale, and batch process — all through a drag-and-drop interface.
Business Use Cases
Product photography, social media graphics, presentation visuals, website heroes, marketing materials, and client proposals — generated in seconds without stock photo subscriptions or designers.
Speech & Voice AI
Convert speech to text and text to speech — all running locally. Transcribe meetings, dictate documents, create voiceovers, and build voice-controlled workflows. Supports Cantonese, Mandarin, English, and 90+ languages.
Whisper Transcription
OpenAI's Whisper model runs locally for real-time and batch transcription. Record meetings, lectures, interviews — get accurate text with timestamps and speaker detection. 100% private.
Text-to-Speech
Generate natural-sounding speech from text using local TTS models. Create voiceovers for presentations, narrate documents, build audio content — with customizable voices and multilingual support.
Voice Workflows
Chain voice AI with your agent — dictate a command, AI transcribes it, executes the task, and speaks the result back. Build voice-controlled automations for hands-free productivity.
AI Coding Assistants
AI-powered development tools that write, review, debug, and refactor code. Whether you're building apps, automations, or data pipelines — AI coding assistants accelerate your development workflow by 2-10x.
Claude Code
Anthropic's agentic coding CLI. Give it a task and it autonomously reads your codebase, plans changes, writes code, runs tests, and creates commits. The most capable AI coding agent available.
Cursor IDE
AI-native code editor built on VS Code. Inline completions, chat-driven editing, codebase-aware suggestions, and multi-file refactoring. Supports Claude, GPT, and local models via Ollama.
Local Code Models
Run code-specialized models like DeepSeek Coder and CodeLlama locally via Ollama. Get intelligent completions, code review, and generation without sending your proprietary code to any cloud service.
AI Safety & Ethics
Running AI responsibly matters. Your MacAI workstation is configured with privacy-first principles, data protection, and ethical AI practices built in — so you can deploy AI confidently for business use.
Data Privacy by Design
All AI processing happens on your hardware. No data sent to cloud APIs, no training on your documents, no third-party access. Your conversations, documents, and business data never leave your office.
Guardrails & Filters
System prompts with safety boundaries, output filtering for sensitive content, PII detection before sharing, and audit logging. Control what your AI can and cannot do with configurable guardrails.
Compliance Ready
Local-only processing helps meet GDPR, PDPO (Hong Kong), and industry-specific data residency requirements. Full audit trail, encrypted storage, and access controls for regulated industries.
How It All Fits Together
┌──────────────────────────────────────────────────────┐
│ YOUR MAC STUDIO │
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ WhatsApp │ │ Telegram │ │ Discord │ Channels │
│ └─────┬────┘ └─────┬────┘ └─────┬────┘ │
│ └────────────┬────────────┘ │
│ ▼ │
│ ┌────────────────┐ │
│ │ OpenClaw │ AI Agent │
│ │ Gateway │ │
│ └───────┬────────┘ │
│ ▼ │
│ ┌──────────────────────────────────────┐ │
│ │ Ollama Runtime │ LLM Engine │
│ │ Llama 3.1 | Mistral | DeepSeek │ │
│ └────────────────┬─────────────────────┘ │
│ ▼ │
│ ┌────────┐ ┌─────────┐ ┌────────┐ ┌──────┐ │
│ │ MCP │ │ RAG │ │ Skills │ │ n8n │ Tools │
│ │Servers │ │Pipeline │ │ Hub │ │Flows │ │
│ └────────┘ └─────────┘ └────────┘ └──────┘ │
│ ▼ │
│ ┌──────────┐ ┌────────┐ ┌──────────┐ │
│ │PostgreSQL│ │ Redis │ │ ChromaDB │ Data Layer │
│ └──────────┘ └────────┘ └──────────┘ │
│ │
│ 100% Local · Zero Cloud Dependency │
└──────────────────────────────────────────────────────┘
All tools are pre-configured on your MacAI workstation. Enterprise tiers include advanced RAG pipelines, custom MCP servers, and dedicated OpenClaw agent configuration. We handle the setup — you use the AI.