← All featuresAI Agent Plan

Review the plan before the agent changes a file.

Use Plan Mode in Flame IDE to have an AI coding agent propose a plan first. Resolve decisions, check approvals, discuss sections, and approve implementation when ready.

The Plan tab with sample project data. Required decisions and approvals stay open until you resolve them, and Approve & Implement waits for you. Select the image to enlarge.
WHY IT HELPS

Settle the open questions while changes are still cheap.

The costliest AI mistakes often start with a quiet assumption: the wrong storage engine, a skipped migration risk, a scope nobody agreed to. Flame's Plan tab turns the agent's proposal into a document you can act on. See the decisions you must make, the risks you must acknowledge, and the steps the agent intends to take, then approve implementation only when the plan says what you want.

Settle decisions before code changes

A plan lists the choices that need a person: required decisions with options, approvals to check, and open questions. The Approve & Implement button stays unavailable until required items are resolved, so the agent starts with answers instead of guesses.

See the risks in plain sight

Plans separate implementation steps, risks and acknowledgements, verification, rollout, and follow-up. You can accept a migration risk deliberately instead of discovering it in the diff.

Discuss it where it lives

Read the plan as an interactive document or as Markdown, search its sections and comments, ask the AI about a section, and compare versions as the plan changes. Feedback stays attached to the section it concerns.

Know what is left to do

A summary bar counts required actions and optional feedback and jumps to the next required item, so you can see at a glance what stands between the plan and approval.

See the architecture after the change

An At a glance card sums up the components, files, decisions, and risks. The plan's interactive architecture diagram shows the system after the change, with new, changed, and removed components marked.

AI AGENT PLAN IN PRACTICE

Answer the questions before they become code.

A good plan is more than a list of steps. The agent's proposal is organized into sections such as implementation steps, decisions, risks, verification, rollout, open questions, and follow-up, and the Plan tab keeps a contents outline beside them.

Decisions with real options

Decisions come first. Each one states the question, one line per option with its consequence, an impact label, and the agent's recommendation with its reason. Pick an option or use the recommendation; it stays marked Needs your choice until you do.

Approvals for the risky parts

Risks come with acknowledgements, such as accepting a migration risk or confirming that you reviewed a backup strategy. Each one is marked until you approve it.

Verification written down

A plan includes a Verification section that states how the change should be checked, such as confirming settings survive a restart. You can read and challenge it before any work begins.

AI AGENT PLAN IN PRACTICE

Review it like a document, not a chat scroll.

Long plans get lost in a conversation. The Plan tab gives the proposal its own place beside your code, with views and tools for reading and revising it.

Interactive, Markdown, Ask AI, Versions

Switch between the interactive plan and its Markdown source. Use Ask AI to question the plan and Versions to see how it changed after your feedback.

Feedback on a section

Leave a comment on the section it concerns, such as asking for a restore path for existing projects. Section feedback shows how many items are open, and plan search covers sections and comments.

An architecture diagram you can explore

Plans can carry a typed architecture diagram with grouped lanes and new, modified, or removed marks. Pan, zoom, and select a component to inspect it. Choose in AI settings whether plans include diagrams automatically, always, or only when you ask. The plan stays a Markdown file you can open in the editor.

AI AGENT PLAN IN PRACTICE

Approval is a decision, not a default.

Starting implementation is the moment the plan stops being a proposal. Flame makes that moment explicit.

Required items gate approval

While required decisions or approvals are open, Approve & Implement stays disabled and the summary bar tells you how many remain.

Overrides are explicit

Open decisions can be left to the agent only through the implementation menu's Approve; AI uses the recommended options. The agent then receives your choices, the recommendation for each open decision, and must state every choice it makes.

Steps you can follow

Implementation steps appear as a checklist inside the plan, so you can see what the agent intends to do and keep track of what remains once work starts.

WATCH IT WORK

See the workflow in a short demo.

Official Flame demo · 0:59

Plan Mode: Review the Architecture Before the AI Agent Writes Any Code

Switch on Plan Mode and the agent explores with read-only tools, then hands you a plan instead of a diff: the scope at a glance, an interactive architecture diagram of the system after the change, and architectural decisions with trade-offs and a recommendation. Nothing is built until you approve.

HOW TO USE IT

From request to approved plan.

  1. Start in Plan Mode

    For a change that deserves review, turn on Plan Mode in AI Chat and describe the task. The agent proposes a plan instead of editing files straight away.

    The Plan Mode switch in the AI Chat composer, turned on before sending a task.
  2. Open the Plan tab

    The plan opens as an editor tab with a contents outline, your decisions and approvals at the top, and each section below. The summary bar shows how many required actions remain.

    Sample plan data. Two required items are still open, so implementation cannot be approved yet.
  3. Resolve decisions and approvals

    Pick an option for each required decision and check each approval. Use Next required to jump to the next open item.

  4. Discuss and revise

    Add comments to sections that need changes, ask the AI about anything unclear, and check Versions after the agent updates the plan. Repeat until the plan matches what you want.

  5. Approve and implement

    Select Approve & Implement when no required item remains. The agent begins the work using the plan you approved.

REQUIREMENTS & LIMITS

Know what the tool depends on.

  • A plan is a proposal that reflects the model's understanding of your request and code. Read it critically; approval does not verify that the plan is correct.
  • Decisions and approvals appear when the plan includes them. A simple plan may have none, and quality varies with the provider and model you choose.
  • Approval starts implementation. It does not confirm that the resulting code is correct, so review the changes before you commit them.
  • Skipping open decisions or unchecked optional items is possible only through an explicit option in the implementation menu. Use it knowingly.
FAQ

AI Agent Plan: common questions