Skip to content
Features

The work moves forward.
The control stays yours.

Focused agents, step-aware routing, and approval before changes. ML.ai Code brings the same ML.ai intelligence into your editor.

Your task → The right scope
ExploreRead only
GeneralCode changes
ArchitectRead only
PlanPlan output
Different agents. Explicit permissions.
Scoped by design

Four agents.
Different responsibilities.

Explore, General, Architect and Plan have distinct jobs. Only General can modify your code.

Read only

Explore

Find and explain repository context without edits.

Code changes

General

Execute a defined task, with changes subject to approval.

Read only

Architect

Lay out the files, trade-offs and order of work.

Plan output

Plan

Write the plan as a document you can inspect.

Approval before action

The diff comes first.
Your permission comes next.

Review the native diff, then Allow, Always or Deny. A denied action stops instead of being attempted through a different tool.

checkout  /  ML.ai CodeLocal example
PayButton.tsx M
12async function handlePay() {
13- await charge();
14 if (pending.current) return;
15 pending.current = true;
16 try {
17 await charge({ idempotencyKey });
18 } finally {
19 pending.current = false;
20 }
21}
Change ready for your review 2 files
fix/checkoutGeneral  ·  Approval requiredReading the task
Beyond read, edit, run

Tools for the work
around the code.

Reusable workflows, independent sessions and connected tools, without turning every task into a new setup.

/commit
Commit messageStart from the current diff
↵
Explain verificationSummarize what the checks proved
Project overviewA packaged workflow

Skills, close at hand.

Use slash commands for packaged workflows, including project and global skills.

Background runs Independent sessions
Trace authenticationExplore
Queued
Prepare checkout fixGeneral
Running
Write migration planPlan
Awaiting approval

Delegate. Keep moving.

Background jobs run in their own sessions and report their progress without needing the panel to stay open.

Command safety
timeout 30 npm test
Wrapper
Executable
Arguments
Inspect the command, not just its first word.

Understand the command.

The safety classifier parses commands and executable positions, including commands inside wrappers.

ML.aiProject / Global
Local Remote OAuth

Connect your tools.

Configure local or remote MCP servers, with project or global scope and authentication where needed.

A deliberate boundary

Your local engine.
Your access token.

Tokens use VS Code Secret Storage. The local engine is separate from the chat interface.

Read the product details
VS Code Secret Storage
Chat interfaceLocal engine
Remote terminalBlocked
Public session publishingBlocked
Engine self-upgradeBlocked

Capabilities summarized from ML.ai Code and ML.ai documentation. The visuals are local product demonstrations.

ML.ai

Know the tools.
Put them to work.

Explore Build and Planning, then start with a task of your own.