Settings reference

ML.ai settings live in VS Code's standard settings UI, plus a set of dedicated ML.ai Settings pages that open as a wide editor view rather than being squeezed into the sidebar.

VS Code settings

Setting	Default	Description
`ml-ai.tier`	`standard`	Which routing tier handles requests: `standard` or `high`. ML.ai decides which underlying model serves each tier; this isn't something you configure directly.
`ml-ai.commandSafety`	`true`	The shell command safety classifier. Turning it off restores prompt-on-every-shell-command behavior; verdicts are still logged.
`ml-ai.enginePath`	(empty)	Path to a custom, non-bundled engine runner, for advanced setups.
`ml-ai.experimental.codeMode`	`false`	The model calls tools by writing a short JS program instead of receiving full tool schemas every turn, saving roughly 1,700 tokens per turn. Each call is still its own activity row and still needs approval. ML.ai's delegation behavior has only been measured with this off. Read at engine startup, so it requires a restart.
`ml-ai.experimental.workspaces`	`false`	Enables worktree/workspace/project-copy/sync surfaces and the isolated-run delegate option. Off by default: these routes touch git worktrees and can move or delete uncommitted work, and some report success in ways that don't match reality.
`ml-ai.denyCredentialAccess`	`false`	Blocks the agent from reading credential files outright.

Model provider

Configure the provider credential ML.ai Code uses for model requests.

ML.ai access token

Signed in. ML.ai is ready to use.

Replace token

Sign out

The wide Settings editor view, Provider page selected

ML.ai Settings pages

Page	What it's for
Provider	Sign in, replace, or sign out of the stored ML.ai access token
MCP	Add and manage MCP servers
Agents	Live catalog of agents from the engine, including custom subagents
Skills	Available skills, selectable for the composer
Advanced	Experimental engine behavior: code mode, worktree workspaces

Command Palette commands

Under the "ML.ai" category:

Command	Effect
Open ML.ai	Reveals the panel
Open Settings	Opens the ML.ai Settings pages
Restart Engine	Restarts the bundled engine process
Copy Engine Diagnostics	Copies a sanitized diagnostics block: engine mode, Node runtime, engine version, host, file-search backend. Never a URL, port, or secret.