ML.ai CodeAgents

General and Explore agents

The two built-in one-turn agents ML.ai ships with, and how they differ from the session's own mode.

Type @ in the composer to select an agent, or pick one from Settings → Agents. The agent catalog is read live from the engine, so it includes any custom subagents you've defined alongside the two built-in ones.

AgentCan edit files?Can run commands?Scope
GeneralYesYesInside the project only
ExploreNoRead-only (search, fetch)Inside the project only

General

A focused, single execution turn. It can edit files and run commands, but only inside the project: it cannot reach outside the workspace, regardless of what the mode's command budget would otherwise allow.

Explore

A read-only search specialist. It reads files, greps, globs, and fetches, and changes nothing. Use it when you want an answer without any risk of a side effect, even one you'd normally approve.

What doesn't change

Both agents hand control back to you after one turn: they don't keep running in the background. Neither one changes the session's standing Build or Plan mode; selecting General inside a Plan-mode session doesn't unlock edits, because General's own scope is a project-only sandbox, not a mode override.

Example

@Explore Find every place this repo reads process.env.DATABASE_URL and summarize the pattern.

@Explore routes the request to the read-only agent regardless of the session's current mode, so this is safe to run even mid-conversation in Build mode: it searches and summarizes, and cannot edit anything while doing it.

For work that should actually change files, use General or the default session agent in Build mode instead.