Plan Mode – Sapience Cloud | Documentation Hub

Use Plan Mode for Amazing Results on Long & Complex Tasks

💡

Equip Plan Mode from the Thinking dropdown, or use Shift + Tab or type /plan. When Sapience uses Plan mode, it uses a rigorous approach (and a complex reasoning model) to lay out all the steps it would take. It will seek your approval, and then march through the steps reliably, leveraging many context engineering tricks to keep the Agent on-task.

Overview

Some work is more than a single reply — it has several steps, touches a few things, and you'd want it done carefully rather than in one hopeful rush. Plan Mode is built for exactly this. Instead of diving straight in, the agent researches, writes a structured plan, and stops to show it to you. Nothing happens until you approve the plan. Then the agent works through it step by step while a live progress bar keeps you in the loop.

It turns "please do this big thing" into "here's my plan — approve it, and watch me tick it off."

How to Turn It On

Three equivalent ways:

WayHowKeyboardPress Shift+Tab in the chat boxToggleFlip the plan switch in the chat inputType itStart your message with /plan — e.g. /plan reorganise my project files and tag each one

Whichever you use, the rest of your message becomes the brief the agent plans from.

What Happens When You Turn It On

The agent does not just start doing the work. Instead:

It researches your request and the relevant context.

It writes a structured plan — with the context, the approach, the steps to take, and how it'll verify the result.

It stops and shows you the plan for approval.

Sapience automatically uses a stronger reasoning model for planning turns, so the plan is genuinely thought through rather than dashed off.

The Plan Itself

The plan is a clear, readable document with a consistent shape — the context, the approach, the files or resources it expects to touch, the tasks as a checklist, and a final verification step. The tasks use tick-boxes, so as the agent works you can see at a glance what's done vs. still to do.

The plan is a saved document attached to your conversation — not a throwaway. You can reopen it any time, even after the work is finished.

The Approval Card — Your Three Choices

When the plan is ready, an approval card appears that you can't miss, with exactly three options:

ChoiceWhat it doesApprove"Looks good — go." The agent begins executing the plan.Reject"Stop." The agent halts, and the plan is closed.Let's chat about thisTell the agent, in your own words, what you'd change. It revises the plan and shows you a new version to approve.

Each choice is just a normal chat reply, so you never leave the flow of the conversation. You can go around "Let's chat about this" as many times as you like — the agent re-proposes an updated plan each time — until it's exactly right, then hit Approve.

The Plan Bar — Watch It Get Done

Once there's a plan on your conversation, a small floating plan bar appears:

It shows the status and live progress — e.g. "3/7 done" — updating as the agent ticks off steps.

The moment you click Approve, it flips to an "Approved" badge, so you get instant feedback (you don't wait for the agent's next turn to see it landed).

It's a draggable panel — grab its header and move it anywhere on screen.

It shows an outline of the plan, and a "More" link opens the full plan document.

The Life of a Plan

A plan moves through a simple, predictable set of states:

(drafting) → Awaiting your approval → In progress → Completed
                     │
              "Let's chat" re-proposes a new version (still awaiting approval)
                     │
                  Rejected

Awaiting your approval — the plan is written and waiting on your decision (this is the approval card).

In progress — you approved it; the agent is working through the tasks.

Completed — every task is done and verification passed.

Rejected — you rejected it; the agent stops.

When a plan reaches Completed or Rejected, it's finished — but it stays saved on your conversation for you to revisit.

What Happens Behind the Scenes

You don't need to know any of this, but here's the shape of it:

When you arm plan mode, the agent gets a plan attached to your conversation and is put on the stronger planning model.

Each turn, the agent is reminded of the current plan and keeps it up to date — ticking off tasks, and updating the plan if something changes (a new step surfaces, or the scope shifts).

The plan you see is always the live one — there's no stale snapshot; the progress and content you read are current as of the last turn.

Availability

Plan Mode works on agents that have planning enabled — most do. If you arm plan mode on a particular agent and nothing seems to happen (no plan, no approval card), that agent simply doesn't have planning turned on. Try it with a general-purpose agent, which will.

Good to Know

Nothing runs without your Approve. The agent never executes a plan until you approve it on the card. You're always the gate.

Turning it on twice is harmless. If plan mode is already active on a conversation, arming it again does nothing — you won't get duplicate plans. (This means a chat that auto-prefixes /plan on every message is safe.)

One plan per conversation. A conversation has a single active plan at a time.

/plan goes at the start. Typed as a slash command, /plan needs to be at the beginning of your message to arm plan mode; buried mid-sentence it won't. (Shift+Tab and the toggle work regardless.)

The plan is yours to keep. It's saved as a document on the conversation — reopen it whenever.

When to Use It — and When Not To

Use Plan Mode when…Skip it when…The task has 3+ distinct stepsIt's a one-line fixIt spans multiple files or areasIt's a single questionYou want a written, approved, tracked record before it's called doneThe agent can finish it in one turn

Plan Mode's overhead — the plan, the approval step, the stronger model — pays off on real multi-step work. For a quick ask, just ask.

Pairs With Brainstorming

If your idea is still fuzzy, start with the Brainstorming skill to shape and approve a design first — then arm Plan Mode to build it. Brainstorming decides what; Plan Mode does it, deliberately.

Summary

Plan Mode turns a multi-step request into a plan you approve before anything runs. Arm it with Shift+Tab, the plan toggle, or /plan; the agent researches and proposes a structured plan; you Approve, Reject, or "Let's chat about this" on an un-ignorable card; and once approved, a floating, draggable plan bar tracks live progress ("N/M done") while the agent ticks off each step. The plan is saved on your conversation to revisit anytime.