AgentOS runs MLX, llama.cpp, LM Studio, and EXO clusters — your code, your models, your machine. No subscriptions. No cloud round-trips. Just an agent loop wired to the models you already trust.
Point AgentOS at a folder. Write project instructions once — stack, conventions, gotchas — and they're injected into every conversation in that project from then on. The model stops re-asking and starts shipping.
.agentos/instructions.md — gitable and yours
Drop reminders, snippets, paths, prompts — anything you want back later. Notes live in a single searchable list, autosave as you type, and stay entirely yours. Nothing flows to the model; the agent never sees them unless you paste them in.
~/Library — diffable, scriptable
Set up a task once and let it run on a cadence — every morning, every hour, overnight. The agent runs headless, logs its work, and drops a notification when it's done. Wake up to your inbox already triaged.
Most local models look great in benchmarks and fall over the first time you ask them to call a tool. We curate the ones that work — Qwen 3.6, Gemma 4, gpt-oss, MiniMax — and ship them as one-click downloads sized for your Mac's RAM.
EXO turns your Mac Studio and your MacBook Pro into a single inference target with shared unified memory. Run models that don't fit on either Mac alone. The cluster view shows live capacity per node and which model is pinned where.
llama.cpp ships bundled and auto-starts. LM Studio plugs in with one click. EXO lets you fan inference across multiple Macs on your network. The Local API Server exposes everything as OpenAI-compatible so Xcode 16 Intelligence and any other client just work. Your active backend stays pinned at the top of the sidebar so you always know what's running.
The Local API Server registers AgentOS as a first-class Intelligence provider in Xcode 16. Add it once in Settings, pick your model, flip tool-calling on — the built-in AI Assistant now runs entirely on your Mac. No cloud round-trip, no per-seat subscription, no code leaving your machine.
AgentOS is a $99 founders-price license while the first 200 spots last — then it goes to $199. No tiers, no subscriptions, no usage caps.
We'll email you when the founders-price gates open. No newsletter, no marketing blasts — just the launch ping.