Build and Plan modes
The two standing modes a session runs in, and exactly what Plan mode does and does not restrict.
ML.ai sessions run in one of two modes at a time. The mode is a standing setting for the conversation: it doesn't change on its own mid-turn.
| Mode | What it does |
|---|---|
| Build | Investigates, edits files, runs commands, verifies the result. The normal working mode. |
| Plan | Investigates and produces a plan. Cannot edit or write files. |
Ask ML.ai Code…
Build
Edits files, runs commands, verifies the result.
Plan
Investigates and returns a plan. Cannot edit or write.
What Plan mode actually restricts
In Plan mode, edit and write tools are switched off outright, not just discouraged by instructions to the model. The command budget is also stricter: commands that write, delete, or execute inline code are refused outright.
Plan mode restricts tools, not consequences. Reading and verification commands (test runners, typecheckers) are still allowed in Plan mode, because verification is part of planning. Running a test suite can still write coverage reports, snapshots, or caches to disk even though no source file changes. "Plan mode changed nothing" is not the same claim as "Plan mode ran nothing."
This is the single most important thing to understand about Plan mode: it is a restriction on which categories of actions the agent may take, enforced by the command safety classifier and the tool set, not a guarantee that the filesystem is untouched by the end of the turn.
Example
Asking ML.ai, in Plan mode, to "fix the registration validation bug and run the relevant tests" will not work as asked: it can investigate and describe a plan, but the edit itself is refused. Asking it to "explain how registration validation works, then propose a fix as a plan" fits Plan mode. Switch to Build mode to have it actually make the change.