Skip to main content

Minimal iOS app

The fastest iOS path in the current preview is Edge Scaffold in a no-model build. It verifies the app shell, signing, Swift package wiring, and real-device build path before you add a model.

Public packages

Edge Scaffold, Edge Kit, Edge Engine, and the Edge Halo binary package resolve from public GitHub and Swift Package Manager URLs. Edge Halo source remains private. The app path still requires Xcode signing, device provisioning, exact version pins, and real-device validation.

Build the minimal app shell​

git clone https://github.com/AtomGradient/edge-scaffold.git
cd edge-scaffold
xcodegen generate
xcodebuild -project EdgeScaffold.xcodeproj \
  -scheme EdgeScaffold \
  -configuration Release \
  -destination 'generic/platform=iOS' \
  SKIP_MODEL_COPY=1 \
  build

SKIP_MODEL_COPY=1 keeps the first build fast and avoids bundling a large model. This build does not prove model quality, learning quality, or App Store readiness. It proves that the iOS app shell and SDK integration compile.

Run on a real device​

Open the generated project:

open EdgeScaffold.xcodeproj

Then:

  1. Select your development team.
  2. Set a unique bundle identifier.
  3. Choose a real iPhone or iPad destination.
  4. Build and run with SKIP_MODEL_COPY=1 for the first launch.

The simulator is not the target runtime for model execution. Use a real device for model load, Neural Imprint restore, memory behavior, and performance validation.

Add the baseline local model​

After the shell runs, configure the baseline model path in edgescaffolding_model_config:

MODEL_COPY=true
MODEL_NAME=Qwen3.5-9B-4bit
MODELS_SOURCE_DIR=$HOME/Documents/mlx-community

Then build without SKIP_MODEL_COPY=1. For larger models, enable the Increased Memory Limit entitlement and validate a Release build on the exact device class you plan to support.

What this app demonstrates​

  • A SwiftUI reference app using Edge Kit and the binary Edge Halo package.
  • Model loading surfaces for local, cached, bundled, ODR, and remote delivery.
  • Neural Imprint restore and correction-learning reference screens.
  • App-owned tool registration and schema snapshot patterns.

Edge Scaffold is not dogfood business logic. Treat it as the smallest current reference app, then replace sample data, sample tools, copy, signing, and model delivery with your app's own policy.

Next step​

After the shell builds, continue to Build the Agent carrier. That guide adds the finance scenario, Edge Studio export flow, model configuration, Neural Imprint restore checks, and real-device validation.