5 parallel coding agents testing SwiftUI: 11 GB → 150 MB RAM, 22 s → 2 s per check. No iOS Simulator (open source)

If you let coding agents work on an iOS app, each one checks its UI changes the slow way: build, boot its own Simulator, install, launch, screenshot. On our production app (~280 Swift files, several big Swift packages, M2 Pro / 32 GB) that's ~22 s per check , and every booted Simulator holds ~2.2 GB and ~210 processes . With 5 agents in parallel my Mac hit a load average of ~1,000 and swapped to a crawl. So I built simless . It runs your real app target natively on Apple Silicon (the "Designed for iPad" slice) as an invisible headless host. One host per git worktree, so every agent gets its own. No Simulator, no windows, nothing stealing focus while you work. What the agent loop looks like $ simless reload --render Welcome.sheet reloaded (live) in 1.9s: compile 1.7s + apply 0.2s Welcome.sheet · light · iphone 402×874 · 25 ms header "Welcome to Notes" #welcome.title text "Your notes stay on this device and sync with iCloud." button "Get Started" #welcome.start @24,779.5 354×36.5 issues (1): ⚠ button "Get Started" tap target 354×36 is smaller than 44×44 Only the edited files get compiled and hot-patched into the running host (~2 s) The screen comes back as a ~150-token accessibility tree , not a screenshot, so you don't burn vision tokens Issues are flagged automatically: unlabeled controls, small tap targets, off-screen and overlapping elements --matrix checks light/dark × iPhone / small iPhone / iPad in one call --png when the agent actually needs pixels simless test runs unit tests on the Mac, also without a Simulator Numbers (Simulator loop = one Simulator per agent, incremental xcodebuild + install + launch + screenshot per check) Simulator loop simless Edit → verified screen, 1 agent 21.9 s Edit → verified screen, 5 agents 37.4 s Memory held by 5 agents 11.2 GB / 1,071 procs Peak load average, 5 cold starts ~1,010 Unit tests (2 classes) 13.1 s The Simulator loop gets slower with every agent you add. simless stays at ~2 s per check. Using it from agents Claude Code: simless skill install once, then just say "use simless for testing" in any project. The skill handles setup, writing fixtures, the edit → reload loop and fixing reported issues. Codex / Cursor / others: paste the 5-line AGENTS.md snippet from the README. Honest caveats It's not pixel-identical to iOS. 8 of 14 demo renders match within 2 pt; List / Form row heights and system accent colors differ. Every screen is in the repo side by side, and simless calibrate runs that comparison for your own app. Still do final checks on a Simulator or device. Benchmarks are a single run on one app. Method and reproduction steps are in docs/benchmarks.md. Requires Apple Silicon, Xcode 26+, an Apple Development certificate and an app-hosted unit-test target. Apache-2.0: github.com/abacusai/simless