Skip to main content
Developer Preview

Run a local model, then validate the learning loop

The first Edge developer path is deliberately small: download the preview baseline model, enter a local multi-turn chat, then use a synthetic sample to generate and restore a Neural Imprint artifact with a local comparison receipt.

The current docs use qwen3.5-9b-4bit as the preview baseline; first download and first load can take time.

Start the CLI pathView the learning demo
Fastest validation path
# install once during Developer Preview
git clone https://github.com/AtomGradient/edge-studio.git
cd edge-studio
python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install -e .

# first local chat
edge models fetch qwen3.5-9b-4bit --source auto
edge demo chat --model qwen3.5-9b-4bit --interactive
Quick start

Three steps: download, chat, learn

Start with a local model chat, then validate the learning loop with a synthetic sample.

1

Install the preview CLI

Use a source checkout during preview. The public pip package is the intended release path, but it is not published yet.

2

Download and chat

Fetch the preview baseline model explicitly, then enter an interactive local chat session before any learning demo.

3

Run the learning loop

Inspect the synthetic sample, generate a local Neural Imprint artifact, restore it under compatibility gates, and compare the receipt.

Integration

Choose your integration path

The CLI path is the fastest way to try Edge. The iOS/Swift path is for app integration and requires preview access plus device validation.

Python / CLI

CLI and local model workflow

Download a model, run a local chat, inspect a synthetic correction sample, generate and restore a Neural Imprint artifact, and compare answer hashes in a local receipt.

Start the CLI pathRead the learning demo
iOS / Swift

iOS app integration

Build the minimal iOS shell, then integrate Edge Kit, Edge Halo, and Edge Scaffold. Requires preview access and real-device validation.

Build the minimal iOS appSwift SDK setup
Build

Core capabilities

Multimodal

Vision

Use vision-language models for image understanding on device.

Vision guide
Voice

Speech and voice

Build local speech-to-text and text-to-speech loops.

Speech guide
Core concept

Neural Imprint

A local personalization artifact that restores user-specific runtime state under compatibility gates without changing model weights.

Model evolution
SDK

Edge Kit

Swift APIs for local inference, model management, EdgeData, EdgeMesh, EdgeSession, EdgeUI, speech, and vision.

Edge Inference API
Local mesh

Device Mesh

Route and transfer artifacts between trusted user-owned Apple devices.

Device Mesh guide