Skip to main content

Developer Preview

Drop this file into your Code Agent

One markdown file. Your AI coding assistant reads it and guides you through everything — from install to deploying a learning agent on your iPhone. Models learn user preferences and tool policy on-device. Data never leaves. Base model stays intact.

Fastest path

Download → Drop into Agent → Go

For your Code Agent
# Download the Agent Guide
curl -O https://atomgradient.github.io/edge-developers/EDGE_AGENT_GUIDE.md

# Give it to your Code Agent and say:
# "Follow EDGE_AGENT_GUIDE.md to help me build a device Agent"

Download EDGE_AGENT_GUIDE.md I prefer reading docs myself

Your Agent will walk you through

Three Aha Moments

1

The model learns user preferences

A finance assistant learns "conservative, cash-flow first." Same model, same question, different answer. Not LoRA. Not prompt stuffing.

2

The model learns tool policy

The Agent learns when to call the cashflow lookup tool and when NOT to call external market tools. Tools are defined by the carrier; policy is learned.

3

It runs on iPhone

The same learning works on your phone. Offline. Local. Removable. The device is the Agent. The app is the carrier.

Compatible Code Agents

Drop EDGE_AGENT_GUIDE.md into any of these

Codex, Claude Code, Cursor, OpenCode, or any Code Agent that reads markdown context.

Prefer to read?

Traditional docs are here too

The Agent Guide covers the main path. Dive into these when you want details.

CLI Learning Demo

Experience the full learning mechanism in your terminal

Device Agent Learning

Experience Agent learning on a real iPhone

Neural Imprint vs LoRA

Why Edge doesn't use fine-tuning or prompt stuffing

Product Architecture

Edge Engine / Kit / Halo / Scaffold / Studio full stack