Skip to main content
Plan Mode turns your agent into a planner. Instead of answering in a single pass, the agent writes a task checklist, then works through it one task at a time — updating each task’s status as it goes until every task is resolved. Use it for multi-step work: research-and-report, “do X then Y then Z”, tool-heavy jobs, or any request where you want the agent to organize the work before doing it.

Enabling Plan Mode

In the Profile step, find the Plan Mode block and turn on Enable Plan Mode. Plan Mode works with or without tools:
  • With MCP servers or Agent-to-Agent — the agent plans, then calls tools task by task.
  • Standalone (no tools) — the agent still plans and tracks a complex, multi-step answer using only its own reasoning.
Plan Mode is available to all users — it does not require Dev Mode. Turning it on runs your agent on the tool-using (native) execution strategy behind the scenes.

Require all tasks completed

Under the toggle, Require all tasks completed before finishing controls how strict the plan is:

Task statuses

As the agent works, each task carries one of four statuses:

Execution limits

When Plan Mode is on, the Execution limits (max steps, max concurrent tools) also apply — Max steps bounds how many iterations the plan loop can run before the agent stops, even without external tools configured.

When to use Plan Mode

  • The request is multi-step and benefits from being organized up front.
  • You want visibility into how the agent is breaking down and progressing through the work.
  • You need the agent to finish everything it set out to do (pair with Require all tasks completed).

Next steps

Back to Agent Overview

Review the full wizard overview and what’s available in each step.