ML.ai Code

ML.ai Code

Reference documentation for ML.ai Code, the VS Code and Cursor extension for agentic coding.

ML.ai Code is a VS Code and Cursor extension that runs a coding agent inside your editor. It reads your repository, edits files, runs commands, and asks before anything irreversible happens. There is no separate server, no CLI, and no login: the agent engine ships bundled inside the extension. It's one of four modules built on the ML.ai Inference harness, and the only one documented here today.

What's in here

This section covers installation, the conversation UI, modes and agents, the permission and safety system, MCP integration, and every VS Code setting ML.ai Code exposes.

FactValue
Extension version0.21.0
Routing tiersML.ai Standard, ML.ai High
Sign-inAn ML.ai access token, no separate API key
PlatformsmacOS Apple Silicon (darwin-arm64), Windows x64 (win32-x64)
Minimum VS Code1.125.0
Credential storageVS Code SecretStorage

ML.ai does not support Linux, Intel Macs, Windows on ARM64, or browser-hosted VS Code in this release. See limitations for the full list and the reasons behind each one.

Start here

How the docs are organized

  • Getting started: install, first run, first prompts.
  • Concepts: the conversation UI, the context meter, Build/Plan modes, reasoning effort.
  • Agents: General and Explore, background runs, isolated worktree runs, custom subagents.
  • Skills & commands: the bundled skills and the slash command catalog.
  • Control & safety: permission prompts, the command safety classifier, auto-approve, diffs and revert.
  • Files & editor: attachments, the file browser, the diagnostics bridge.
  • MCP: adding and managing MCP servers.
  • Settings reference, Troubleshooting, Security & privacy: standalone reference pages.
  • Reference: a full limitations table and the changelog.