Private local AI, set up properly on your Mac.

Safety First on the Frontline

Some "Private AI" claims ask you to take their word for it. The Ambassade Protocol doesn't; it's a clear account structure — inspectable, so privacy and security aren't just settings you simply hope stay on.

• Our 'Diplomatic Immunity' account structure and naming convention — Ambassador, Attaché, Chancery, SCIF — enforces the benefits of the Protocol at the OS level, as is fully explained at config time. Account names can always be changed and added, of course, to suit other naming choices if desired
• Two default Safe types for your most sensitive material: the High-Security Personal Safe (HSPS™) for you alone, and the High-Security Mission Safe (HSMS™) for shared work
• For those who need it: a fully sealed SCIF companion account per role, enforced with kernel-level network isolation — a wall built into macOS itself, not a setting that can be quietly switched off
• A four-layer hierarchy — Outpost → Sealed-Chamber → Strongroom → Safe → Files — nests your most sensitive material behind OS-level locks, not just folder-naming conventions

Five Private, Safe AI Portals

Every conversation with your local AI needs a conduit between you and the model; some are 'agentic' and others 'simple' chat. We install and configure several for you, suited to many kinds of work and play:

• OpenClaw and Hermes Agent — for agentic work: researching, writing, executing, coding, creative endeavors; they can work autonomously 24/7 on almost anything you like. Both come with excellent instructions/web resources
• Ollama — the runtime engine underneath every channel, serving your local LLMs and SLMs. An Ollama App lets you query models directly, bypassing the latency of the Agentic use of the very same models
• LM Studio and oMLX — sandboxed “tasting rooms” for evaluating new models before you deploy them, with tools to test context size, timeouts, reasoning behavior, vision, audio, OCR, translation, and more.
• Optional installs if desired: Obsidian as a private 'second brain,' Datasette as a full conversation-archiving tool, and Open WebUI for a familiar browser chat window (also a frontend to OpenClaw and Hermes Agent if you so choose)

LLMs Paired for YOUR Mac

A model that can technically be installed on your Mac is not necessarily a model you'll enjoy using. Fit matters more than specs. We recommend LLMs (and SLMs - Small Language Models) that make sense.

• 'Local Duty Officers' (Gemma, Mistral, Llama, Qwen, DeepSeek, Ornith, etc.) serve free of charge (except for electricity charges) 24/7/365, by default (primary local model always loads first)
• 'Visiting Dignitaries' (Claude, ChatGPT, Gemini, etc.) are invited in only when you want extra capability, billed per token or by subscription (while local AI is free)
• Speed vs. usability tradeoff review
• Full agentic use, or simple 'Local ChatGPT' style use — your call
• We've found traps so you don't have to: some quantization levels quietly fail at vision tasks that others models handle perfectly. We know which is which before we recommend anything
• Honest performance expectations: First response times on modest hardware, the genuine privacy of local AI vs cloud speeds tradeoffs, keeping local AI models 'warm', and hence faster. We'll show you how (and share YouTube links)

A good fit

• Consultants, writers, researchers, founders, lawyers, creatives, small-business operators, office managers, students and homemakers
• Professionals with confidential documents on their Macs
• Mac users who bought an M-series Mac for use with local AI
• Those who have downloaded OpenClaw, Hermes Agent, Ollama, LM Studio, Obsidian, Open WebUI or adjacent tools and find themselves stuck on uncertain about privacy and security

Built to Grow With You

• This is a careful, hands-on process — not an overnight miracle. We'll tell you plainly what a setup can and can't do.
• Every account can include a sealed, SCIF-style compartment — its own isolated space for anything that should never touch the network, enforced at the OS level.
• Pricing reflects real, hands-on time. We'll always explain what you're paying for before we start.