Skip to main content
Plan Mode adds a review and approval layer before Purna begins a complex task. Instead of moving directly from your request to execution, Purna clarifies the scope, creates a detailed plan, and waits for your approval. Use Plan Mode when a task has multiple steps, requires specialized tools or compute, or needs clearly defined outputs and validation criteria.

​
1. Submit and clarify your request

Start with the scientific or technical outcome you want to achieve. For example:
If the request leaves important choices open, Purna asks follow-up questions before creating the plan. These questions establish the task scope and prevent the execution agent from making avoidable assumptions. Include relevant files, preferred methods, required formats, and validation expectations when you already know them.

​
2. Review the proposed plan

Purna translates the clarified request into an executable roadmap. The plan separates the work into explicit sections so you can review both the intended result and how Purna proposes to reach it.
Plan Mode approval screen showing capabilities, assumptions, constraints, validation, and overall success criteria

Review the proposed approach, deliverables, validation checks, capabilities, assumptions, constraints, and success criteria.

Some capabilities use external providers or consume additional compute. When approval is required, the plan identifies the provider and operation before any compute is spent.

​
3. Approve or revise the plan

Choose the next action after reviewing the roadmap:
  • Click Approve to lock the current plan version and begin execution.
  • Click Request changes to revise the scope or approach before work begins.
  • Click Cancel to stop the task.
After approval, the locked plan appears in Approved roadmap on the right panel. Purna passes that approved version to the appropriate execution or research agent, creating a clear boundary between planning and execution.
Plan Mode does not automatically turn every task into a Deep Research run. It can prepare many kinds of multi-step work, including literature research, data analysis, code execution, molecular modeling, and artifact generation. For systematic literature reviews, see Deep Research.

​
4. Follow the execution

You can open the execution timeline in either of two ways: click View reasoning on the task card in the chat, or open Compute in the right panel and select the running task. Both routes open the execution trace, where you can follow the approved roadmap alongside the work being performed. The execution layer records:
  • The active agent instance and task status
  • A chronological timeline of execution steps
  • Data sources, tools, scripts, and specialized providers used
  • Inputs, outputs, and tool results available for inspection
  • Pauses, approvals, review decisions, and failures that require attention
  • Files created while the task runs
Plan Mode execution view showing a live timeline, tool calls, and the approved roadmap

Follow the approved roadmap and inspect tool activity in the live execution timeline.

You can continue working elsewhere in Purna while the task runs. Return through View reasoning or Compute whenever you want to check progress or inspect a step in more detail.

​
5. Review generated files

When the task produces an artifact, it appears in the session’s Files section. The exact file types depend on your request. A computational biology workflow may generate:
  • CIF structural files (.cif)
  • Structured metadata and metrics (.json)
  • Reports and explanations (.md)
  • Reproducible analysis or generation scripts (.py)
Purna also creates answer.md when the workflow includes a final written summary. It consolidates the key findings and points you to relevant information across the other deliverables.
Files section showing generated Python, JSON, Markdown, CIF, and answer.md outputs

Open or download generated reports, scripts, metadata, structures, and the final answer from Files.

Supported scientific outputs can be opened directly in Purna. For example, you can inspect a generated CIF structure in the built-in molecular viewer without leaving the session.
Generated GLP-2 and GLP-2R complex displayed in the molecular structure viewer

Inspect a generated protein complex directly in Purna's molecular structure viewer.

​
6. Continue with follow-up requests

After the task finishes, you can continue working in the same session. Ask a related question, request a revision, add another validation step, or build a new output from the completed work. Purna can use the existing session context, including:
  • Your original request and clarifications
  • The approved plan and earlier decisions
  • The execution timeline and tool results
  • answer.md and the other generated files
For example, you can ask Purna to compare an additional structure, rerun an analysis with different constraints, revise the final report, or create a visualization from the generated data. You do not need to upload the same files or restate the entire task. If a follow-up substantially changes the scope or requires new tools or compute, Purna may prepare another plan for your approval before continuing.

​
Watch Plan Mode in action

See how Plan Mode clarifies, plans, and runs a multi-omics data analysis task.

Plan Mode in Purna: running a multi-omics data analysis job