code.ae

The coding agent that checks its own work.

It plans, writes the change, runs your tests and opens your app in a browser. Then it tells you what it saw.

code.ae DesktopInteractive demo
  1. Your idea
  2. Agent at work
  3. The result

An idea becomes a website.

Your idea

Build a coffee shop website. Warm colours, a simple menu, and prices in AED.

Illustrative walkthrough

A real session, from prompt to a working page.

Three modes. You choose.

Switch any time with one command.

  • /plan

    Plan

    Reads your code and proposes numbered steps. Changes nothing until you approve.

  • /build

    Build

    Makes the change, runs it, and fixes what fails. Splits big jobs across helpers that work in parallel.

  • /design

    Design

    Agrees the screens with you first: layout, states, and one main action per screen.

code.ae Plan modeInteractive demo
  1. Your approval
  2. Agent at work
  3. The result

You decide what happens next.

Your idea

Build a team chat app with channels and a clean, simple interface.

  1. 1Create channels and conversations
  2. 2Add a message composer
  3. 3Check the chat flow

Review the plan. Nothing starts until you approve.

Illustrative walkthrough

It does not say done until it is.

Every change goes through the same four steps.

  1. Reproduce
  2. Change
  3. Run tests
  4. Check the page
Still red? Back to change.

Agree the design before the code.

Design Studio shows each screen as an outline. You approve it, the agent builds to it, then checks the result against it.

Pick a model. Or bring your own.

The same agent runs on every model below.

ModelSolvedPer task
DeepSeek V4.1 FlashDefault10/10
Qwen 3.8 Flash10/10
MiMo V2.6 Flash8/10
Laguna S 2.18/10
Ling 3.0 Flash7/10
Kimi K2.7 CodePremium7/1027¢

10 real bug-fix tasks per model, same agent and settings. Method on the benchmarks page.

Your own key

Point it at any OpenAI-compatible endpoint. Your key stays on your machine.

code-ae provider add mine --endpoint <url> --model <id>

Every token counts.

Most of what the agent re-reads is served from cache, which costs about a tenth.

From cache 88%New 12%
  • per solved task on the default model
  • 0tokens spent tidying long sessions
  • each file is sent once, never twice

In your terminal and your editor.

The CLI and the editor extension run the same agent and share sessions and project notes.

code.ae CLIInteractive demo
  1. Your idea
  2. Agent at work
  3. The result

One prompt. A working checkout.

> code-ae

Add a checkout with a cart summary, delivery details, and order confirmation.

Illustrative walkthrough

You stay in charge.

  • Review every edit before it lands.
  • Undo any turn from a checkpoint.
  • Plan mode cannot change a file.
  • Project notes live in a file you can read.