Navigation
ELOWEN · Documentation

Last updated: October 27, 2026

Browse documentation · What Plan mode allows
Autonomous work and permissions

What Plan mode allows

Plan mode is a permission guardrail: it lets Chetty investigate and write a plan, and refuses anything that would change your work.

What planning can do

While planning, Chetty can use any tool that only reads. That includes searching files, listing processes, reading sub-agent results and looking up plugin data. It can also keep its own task list.

Anything you delegate from a planning turn runs read-only, so a child cannot make changes on the plan's behalf.

Tip: Ask for the investigation you actually want, such as "compare both approaches and tell me which fits the current tests". Planning rewards questions about evidence, not requests to fix things.

What Plan mode refuses

  • Edit and Write can change only the session's plan file.
  • Bash runs only non-destructive commands.
  • Every tool that could change your files, data, processes, messages or external services is refused.
  • Tools from connected MCP servers are refused, because Chetty cannot tell which of them only read.

Plan mode is a guardrail, not a sandbox: it limits which tools a turn may use, not what your account is allowed to do.

A read-only limit does not leak out of the reply it belongs to: a follow-up to an existing sub-agent runs read-only for that reply, and the child keeps its own access for later replies outside Plan mode.

Finish the plan

When the plan is ready, choose Implement plan to accept it, switch the conversation to Build and start work. Choose Keep refining to stay in Plan mode and keep working on the plan.

Accepting a plan does not bypass permissions: the same rules apply once implementation starts. For the step-by-step flow, read Plan, work, and approve the next step.

Admin: Where sub-agents are available, Plan mode suggests read-only explore sub-agents to study the code and plan sub-agents to propose designs in parallel. Plan mode: explore agents and Plan mode: plan agents in Settings → <assistant name> AI → Runtime cap how many of each are suggested (0–5, default 3); at 0, Chetty does that step itself.

In a managed Project

In a managed Project, the plan stays inside that Project's environment. File searches and non-destructive shell commands work during planning, including for read-only sub-agents.

Only the planning conversation can edit its plan. Clearing the conversation also clears the plan before planning resumes.