Local model gateway
An OpenAI-compatible endpoint over local GGUF models via llama.cpp — point any OpenAI SDK at localhost and go.
Local-first · Apple Silicon first · Open source
The local-first AI OS for developers
Your models, your machine, your keys. Maya OS runs local LLMs, autonomous agent teams, and a hardware-aware model recommender — all behind one command.
npm i -g maya-os
maya serve
npm i -g @maya-os/code
mayac
Maya OS · maya-os
An AI OS that lives on your Mac
One command boots a local model gateway, a team of autonomous agents, and a dashboard to drive them — no cloud account required.
RAM guardrails + hardware-aware fit
Maya knows your machine. It sizes models to your RAM so you never swap-thrash or OOM a load.
Agent teams
Real-PTY standing agents — Claude Code, Codex, or maya code — talk on a shared channel and build things autonomously.
Web dashboard
Chat with your agent teams and models, browse files and runs — all from a local web UI.
Hardware-aware model recommender
Detects your Mac and ranks local models by fit, quality, and speed — so you download the right one the first time.
qwen2.5-coder-7b
perfect
deepseek-coder-lite
good
deepseek-coder-33b
too-tight
npm i -g maya-os
maya code · @maya-os/code · bin: mayac
The coding agent that runs on your terms
A full-screen terminal coding agent — provider-agnostic, subscription-or-key, and extensible. This site was built by maya code agents.
Provider-agnostic
Anthropic and OpenAI, side by side. Run on a plain API key or your own subscription — your call.
Full-screen TUI
A fast, keyboard-driven terminal interface for real work — read, edit, run, and review in one place.
Skills
Load reusable, skills.sh-compatible skills on demand — procedural knowledge the agent pulls in when a task matches.
MCP
Connect Model Context Protocol servers to give the agent new tools and data sources at runtime.
Sub-agents + workflows
Fan out well-scoped work to parallel sub-agents and orchestrate multi-step workflows.
Memory + background tasks
Durable project memory across sessions, plus long-running background jobs that keep working while you do.
npm i -g @maya-os/code
See it run
Left: the gateway + recommender at boot. Right: maya code driving a real task.
$ maya serve
Maya OS ▸ local-first AI OS
──────────────────────────────────────
✓ Detected Apple M2 Pro · 16 GB
✓ Gateway 127.0.0.1:4100/v1
✓ Dashboard 127.0.0.1:4200
Model recommender — for your Mac
──────────────────────────────────────
● perfect qwen2.5-coder-7b Q5_K_M
● good deepseek-coder-lite Q3_K_M
○ too-tight deepseek-coder-33b Q4_K_M
RAM guardrail: loads capped for safety
Ready. Point any OpenAI client here →maya code · claude-opus-4-8 · sub
› build the landing hero, wire the copy
● read COPY.md
● edit index.html (+42 −3)
● bash python3 -m http.server 8080
✓ agents design · build · content · deploy
skills · mcp · sub-agents · memory
──────────────────────────────────────
⏎ send ⇥ tools ^K cmds ^C quitReal output shapes from maya serve and the mayac TUI.
Why local-first
Your models, your machine, your subscription. Nothing leaves your box unless you point it at a cloud provider.
Private by default
Prompts, code, and context stay on your Mac. No telemetry pipe to someone else's servers — you own the whole loop.
No metered surprises
Local models cost you nothing per token. Bring a cloud key only when you want to, only where you want to.
Yours, offline-capable
Once a model's on disk, you keep working — on a plane, behind a firewall, or off the grid.
Cloud-optional, never cloud-required.
Boot your own AI OS in one command
Free, open source, Apple Silicon first.
npm i -g maya-os
maya serve
npm i -g @maya-os/code
mayac