Arch™

A native macOS coding agent for any project, with first-class Swift and Xcode workflows. Choose a local or cloud model, then hand it a task.

How it works

Four providers, your choice — Ollama (local, free), Claude, OpenAI / Codex, or Apple Intelligence (on-device, Apple Silicon only). No forced subscription. Switch per conversation.
Coding first, Xcode fluent — point Arch at any project folder. Swift projects get scheme discovery, structured test results, Xcode-aware guidance, and an Open in Xcode action; other stacks keep the same project-scoped tool loop.
Built to keep going — an encrypted job queue survives relaunch, retries provider failures, resumes after login, and protects each job with project-specific command rules and a pre-work Git checkpoint.
Private by construction — conversations stay on your Mac. Cloud-provider messages go directly to the provider using your own key. Ollama and Apple Intelligence can stay on-device. Nothing passes through a server I control.
System prompt per conversation — describe your project, its conventions, and what the model should never do. It sticks to that conversation.
Parameter control — temperature, Top P, time and tool-round budgets, command policy, credential access, and Git checkpointing are explicit rather than hidden.

Chat. Task. Done.

What it will never do

No analytics, no ad SDKs, no tracking. Conversations don't flow through a server I control. File tools are constrained to the project folder you pick, including across symlinks. Shell commands require a separate opt-in because commands inherit the access macOS grants to Arch. Archie access is also off until you enable it.

Requirements

macOS 26 or later. Apple Intelligence is the built-in, no-install local route on supported Apple Silicon Macs. Ollama is optional and must be installed separately only if you want to use Ollama-managed models.