Skip to content

Feature request: Persist agent thinking/reasoning output for review after response completes #9702

@DocterJac

Description

@DocterJac

Summary

Feature request to persist agent thinking/reasoning output so users can review it after the response completes, rather than losing it when the transient thinking preview disappears.

Problem

When the agent processes a request, its thinking/reasoning output is displayed as a transient streaming preview. Once the agent finishes responding, the thinking text disappears and cannot be reviewed. This is a significant friction point for users who rely on the thinking output to:

  • Understand agent decision-making — why the agent chose a particular approach over alternatives.
  • Refine and improve prompts — seeing how the agent interprets requests helps users craft better prompts.
  • Debug unexpected behavior — when the agent does something unexpected, the thinking output is the primary window into what went wrong.
  • Learn the agent's interpretation model — understanding how requests are parsed and acted on.

This affects every session. The thinking output is paid-for content that provides genuine value, but its transient nature makes it nearly impossible to digest in real time, especially for longer reasoning chains.

Reproduction steps or desired workflow

  1. Send a non-trivial prompt to the agent.
  2. Observe the thinking/reasoning output as it streams.
  3. After the agent completes its response, the thinking output should remain accessible rather than disappearing.

Desired capabilities (in priority order):

  1. Persistent expandable/collapsible section — Each agent response should include a toggleable "Thinking" section that preserves the full reasoning text. Collapsed by default to keep the UI clean, but expandable on demand.
  2. Dedicated Thinking Log panel or sidebar — A separate view that aggregates all thinking output from the conversation, making it easy to review the agent's reasoning across multiple turns.
  3. Export with thinking included — When exporting or sharing a conversation, include the thinking output so it can be reviewed offline or shared with collaborators.

Artifacts

None attached.

Warp version

v0.2026.04.27.15.32.stable_03

Operating system

macOS

Metadata

Metadata

Assignees

No one assigned

    Labels

    area:agentAgent workflows, conversations, prompts, cloud mode, and AI-specific UI.enhancementNew feature or request.needs-infoMore issue-specific reporter detail is needed before the problem can be confidently triaged.os:macmacOS-specific behavior, regressions, or requests.repro:mediumThe report suggests a plausible repro path, but some uncertainty remains.triagedIssue has received an initial automated triage pass.

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions