Setting to keep thinking and tool activity expanded by default

[krishhgg]

Problem

Nightly now shows thinking traces (#11784). The row opens while the model thinks, then closes when that burst finishes. To read it you click the thinking icon every time.

Tool/work folds do the same. Worked for … is the collapsed summary of tools the agent ran. Following the full chain, thinking, tools, then the answer, means opening rows by hand all turn.

This is not discussion #6756. That one remembers a fold after you switch threads. This one changes the default: keep the transcript open unless I close a row myself.

Proposal

A persisted client setting, like Conductor’s “Don’t collapse tool calls (Garry Mode)”, Conductor’s toggle that leaves tool output open instead of auto-collapsing it.

• Off (default): current behavior. Thinking opens while live, then closes. Tool/work groups stay folded.
• On: thinking rows and tool/work groups stay open by default, including after the turn settles, the turn is done and no longer streaming.
• Clicking a row still toggles just that row.
• A keybinding toggles the setting. Conductor uses Ctrl+O. T3 already uses mod+o (Cmd+O on Mac, Ctrl+O on Windows/Linux) for Open Favorite, so don’t reuse that. Pick a free shortcut, or leave it unbound until set in Settings → Keybindings.

Scope: web/desktop chat timeline. Reasoning rows, tool-call rows, and the Worked for … work fold. Not diffs, those already have their own default.

Why

People debugging agent decisions need the thought + tool trail visible without hunting. Compact is the right default for most people. This should be opt-in so the dense view doesn’t become everyone’s problem.

Smallest useful scope

One boolean in client settings, a Settings → General row, and use that boolean as the default for reasoning disclosures and work/tool folds. Per-row toggles still win on top of that default.

Alternatives considered

• Reasoning-only expand setting. Incomplete, because the next click is still the tool group.
• Comment on [Feature]: Preserve reasoning/tool trace expanded state when switching threads #6756. Different request: persist state vs change the default.

[asgeirtj]

+1, and a receipt that the "smallest useful scope" here is accurate. On current main, the collapsed default is hardcoded in the timeline builder: apps/web/src/components/chat/MessagesTimeline.logic.ts constructs every activity/work group with

expanded: input.expandedWorkGroupIds?.has(groupId) ?? false

(four call sites). The per-row toggle state already flows through expandedWorkGroupIds, so the proposed boolean only needs to feed that ?? false fallback. Per-row clicks would still win on top, exactly as proposed. No new state plumbing needed.

[caitlon]

+1, and a suggestion for the shape: a per-thread Transcript view with three levels instead of one global on/off, the way the Claude desktop app's Code mode does it (thread menu → Transcript view → Normal / Thinking / Verbose):

• Normal: today's behavior. Thinking opens while live and folds after, tool groups stay under Worked for ….
• Thinking: thinking traces stay expanded after the turn settles; tool groups stay folded.
• Verbose: thinking and every tool call with its output stay expanded.

Per thread matters because the need differs by thread: a debugging thread is worth reading in full, a long refactor is not. A global default plus a per-thread override would cover both this proposal's setting and the case above, and the keybinding could cycle the three levels.