What each setting does, what it starts as, and what to set it to.
Theo’s pick is something he says in a video. From the docs is what T3 Code’s own docs recommend. My suggestion is my reading of the docs, not Theo’s view.
Threads
Working section
Shows a Working section in the sidebar. Turn it off in Settings → General. Each device keeps its own choice.
Default
On
Theo’s pick
OnIt hides threads while they're actively working, so only threads that need attention are in your face. Theo says he can't live without it now.
Sources
18:42That's cuz we have a working section on the bottom. This is optional. I currently have it as a beta feature, the working section. This hides threads while they are actively working. I thought this was going to be another one of those changes people hated. I can't live without it now and I've seen a lot of other people doing the same.
Controls how tool calls appear in chat; the stored options are showAll, hideSuccessful, hideAll and zen. It is set in Settings → Chat (web and desktop) or Settings → Appearance and applies to this browser or device only.
Default
Show all
My suggestion
Hide successfulThis keeps long agent runs readable. Because the setting applies to this browser or device only, you can keep a different choice on another device.
Sets the choices in the thread snooze menu. It is under Settings → Threads → Snooze options.
Default
Not stated in the docs
My suggestion
Match your working hours, e.g. tomorrow at your start timePresets can wake after a fixed time or at a local time on a given day, with up to 8 entries. Aligning them with your schedule makes snoozed threads return when you can act on them.
Custom instructions in Settings → Threads that change how automatic thread titles are written.
Default
Not stated in the docs
My suggestion
A short style rule, e.g. "Start with the ticket ID, max 6 words"The text generation model writes titles, so short instructions shape every title it writes. This keeps the sidebar consistent and easy to search.
A thread behavior setting that snoozes threads by default when they stop on a usage limit. Providers without a reset time offer a manual option instead.
Default
Off
My suggestion
On if you hit provider limits oftenEnabling it in thread behavior settings snoozes limit stops by default.
A settle rule (sidebarAutoSettleOnMerge) among the thread settling rules in Settings → Threads. Those rules also include settling inactive threads after seven days by default.
Default
On
My suggestion
Off if you keep working in threads after their PR mergesSettling on merge is a separate settle rule (sidebarAutoSettleOnMerge) that you can change in Settings → Threads.
Decides whether messages sent while a turn runs are queued for a later turn or steer the running turn immediately. A shortcut performs the opposite action.
Default
Steer
Theo’s pick
SteerTheo doesn't like queuing, so he sets follow-ups to steer the running agent instead of queuing a message.
Sources
18:05…have steering too, so when the model's annoying you, you can send a message to do a follow-up. We also have queuing, so queued message, and you can have a hot key to swap between the two. I don't like queuing, so I set my default to be steer, but you can change that in settings.
In Settings → Chat → Send shortcut, choose how sending works; the options are enter, mod-enter-multiline and mod-enter.
Default
Enter
My suggestion
Cmd+Enter (Ctrl+Enter on Windows/Linux) for multiline promptsRequiring a modifier to send makes it less likely you send a half-written prompt; the setting offers mod-enter-multiline and mod-enter options.
Settings controls the capture sound, the brief flash on the captured window, and the capture animation. The system reduced-motion setting also disables the animation.
Default
On
My suggestion
Off in shared or quiet spacesThe flash still confirms the capture visually, so turning off the sound loses little feedback.
Dictation shortcut in Voice settings picks how the keybinding behaves; the modes are auto, hold and toggle, and Toggle always needs a second press. Change the chord itself in Keybindings settings.
Default
Auto
My suggestion
Toggle for long dictationToggle always needs a second press to stop, so you don't have to hold keys through a long message.
Your chosen transcription environment skips non-speech audio.
Default
On
My suggestion
Off if quiet speech gets droppedFiltering skips non-speech audio. If soft or distant speech is being cut, turning it off sends everything for transcription.
A preferred microphone order in Settings → Voice. Voice input uses the first connected microphone in the list.
Default
Not stated in the docs
My suggestion
Put your best headset or mic firstRecording uses the first connected entry, so ordering a dedicated mic first means it is picked whenever it is plugged in.
A Labs beta. After a Claude or Codex turn completes, the agent predicts your next prompt.
Default
Off
My suggestion
On if you mostly use Claude or CodexOnly Claude and Codex predict prompts. Codex follows a change at once, but open Claude threads pick it up only after they restart.
In Settings → Voice → Transcription, download a model and choose Use model.
Default
Parakeet unified English 0.6B
My suggestion
A multilingual model if you dictate in languages other than EnglishThe default model is English-only. Choosing a multilingual model helps if you dictate in other languages.
A switch in Settings → Maintenance that controls whether T3 Code checks for newer provider versions. Settings → Providers shows those updates for the selected environment.
Default
On
My suggestion
Off only on hosts where provider versions are pinned or centrally managedWhen checks are off, refreshing will not find newer versions until you turn them back on. Keep them on wherever you update providers yourself.
Settings → Threads → Text generation model picks the model that writes generated text. Settings → Source control & issues and Settings → Voice can each pick a different one. By default, a coding agent such as Codex or Claude writes this text.
Default
Codex
My suggestion
A dedicated generator added under Providers → Text generatorsBy default, a coding agent writes this text and starts a session for each piece. You can add a generator for it under Settings → Providers → Text generators.
How many days the server keeps scanned usage history. Set it as usageHistoryRetentionDays in userdata/settings.json. While settings.json cannot be read, nothing is pruned.
Default
92 days
My suggestion
365 days if you track yearly costClaude Code deletes its own history, so the server's scanned copy can be the only long-term record. Raising the number keeps a longer history.
In Settings → Providers → Advanced, controls how idle Claude threads over 200,000 tokens are compacted. Each Claude provider can override it. When you return shows a Compact chip.
Default
When you return
My suggestion
Before cache expires, for long Claude sessions you revisitCompact idle threads in Settings → Providers → Advanced controls idle compaction for Claude threads, and before cache expires is one of its modes.
Sets the permission mode for new threads in Settings → Threads → New threads → Permissions. Projects can override the environment default.
Default
Full access
Theo’s pick
Auto (or Full access if you're not being careful)Theo says supervised and accept-edits modes are tools from the past. You should use auto if you're careful and full access if you're not.
Sources
40:21Like, supervise and auto accept edits are tools from the past, from like 2024 even at like newest. Use auto if you're careful and use full if you're not, period.
Removes recognized agent co-author credits when merging. It is in Settings → Source control & issues, and projects can override it.
Default
Off
My suggestion
On if your repo forbids bot co-authorsIt removes only recognized agent credits; human co-authors stay credited. A project override lets you apply it only where it is required.
Cleanup policies in Settings → Worktrees → Cleanup can be enabled on one machine or on all connected environments. They run on the server. A worktree is a separate copy of your project's files so agents can work without clashing.
Default
Off
From the docs
OnThe FAQ tells users to turn on automatic worktree cleanup because worktrees add up fast. One user cleared 500 GB.
Keeps the default-branch checkout up to date with its configured upstream. Set it per environment or override it per project in Source control & issues.
Default
Not stated in the docs
My suggestion
OnIt keeps the default-branch checkout up to date with its configured upstream, so new work starts from a current default branch.
Sets an absolute path, such as D:\worktrees or ~/worktrees, where new worktrees are created on this machine. It is per machine; existing worktrees stay where they are, and cleanup covers both the default folder and the custom one.
Default
Not stated in the docs
My suggestion
A folder on a roomy drive, e.g. ~/worktrees on a large diskWorktrees add up fast, and one user cleared 500 GB. Moving them to another drive keeps the system disk free, and cleanup still covers the old default folder.
A Labs beta turned on per Mac in Settings → Labs. New worktrees are cloned from a cached checkout instead of writing every file.
Default
Off
My suggestion
On for Macs that create many worktrees in large repositoriesIt is macOS-only and works per Mac. Cloning from a cached checkout avoids writing every file, which helps most in big repositories. It is still a beta.
The static prefix put in front of automatically named new worktree branches, set in Settings → Worktrees → Worktree branch naming.
Default
t3
My suggestion
Your team's convention, e.g. your initialsAutomatically named worktree branches use this prefix. Matching your team's branch convention makes agent branches easy to recognize on shared remotes.
A beta switch in Settings → Source control & issues. It adds review guides, which the review guide model writes when you choose Write guide.
Default
Off
My suggestion
On if you review agent-written pull requestsWrite guide has the review guide model read the change and produce a guide, which helps you review large agent diffs. It is still a beta.
Chooses which model writes review guides, set in Settings → Source control & issues → Review guide model.
Default
Not stated in the docs
My suggestion
A strong reasoning model you already have set upThis model reads the change to write the guide. Picking one explicitly avoids depending on whichever provider happens to be set up.
In Settings → Threads (Source control on mobile), lays clones out like ghq, so they default to the base directory followed by host/owner/repo.
Default
Off
My suggestion
On if you clone many repos or already use ghqClones then land in host/owner/repo folders under your base directory, so repositories with the same name from different owners do not collide.
In Settings → Worktrees under Advanced, controls how new worktrees initialize submodules. A project can set it too, and otherwise a t3.json value from the checked-out branch can supply it.
Default
Inherit (t3.json, then recursive)
My suggestion
Set explicitly for repos with many nested submodulesThe docs point repositories that declare many nested submodules to this setting, found in Settings → Worktrees under Advanced.
In Source control & issues, sets the default pull request merge method.
Default
Not stated in the docs
My suggestion
SquashMerge cleanup still recognizes squash merges on GitHub when the pull request targeted the default branch, and squash keeps one commit per agent task.
Per-provider switches under Settings → Source control & issues that choose which hosts Add Project and Publish Repository offer.
Default
Default host only
My suggestion
Turn on each host you actually clone from or publish toHosts you have not set stay off, except the default one. Turning on the hosts you use keeps the pickers short but complete.
In Settings → Source control & issues → Commits & pull requests, choose a source control writing style, such as Repository conventions.
Default
Not stated in the docs
My suggestion
Repository conventionsRepository conventions is one of the styles offered in Commits & pull requests. Matching the repository's existing style makes generated commits and pull requests blend in.
Automatically chooses a machine for new work across your environments. It is under Settings → Environments → Load balancing on web and desktop.
Default
Off
Theo’s pick
On (auto balance across connected servers)If you don't want to pick a server for each thread, auto load balancing spreads new threads across all the machines you have connected to T3 Code.
Sources
44:40And if you're too lazy to even pick which server to run, Maria has built an awesome feature for you in settings here. Load balancing. You can now set up auto load balancing across all of the servers that you have connected in T3 code, so it will fire off the threads across your different machines.
With the T3 Connect settings in Settings → Sharing & access on the host, makes each new device signed in to your T3 Connect account wait for approval from a device that is already connected.
Default
Off
My suggestion
OnWithout it, any device signed in to your T3 Connect account can open a linked machine. With approval on, a new device key waits until someone already trusted approves it.
A beta in Settings → Labs for the machine that runs the threads. It allows guest links. While it is off, every guest link on that machine stops working.
Default
Off
My suggestion
On only while you are actively sharing threads with teammatesTurning it off refuses new links and pauses existing ones. It is a single switch that stops all guest access on the machine.
A beta in Settings → Labs that links your machines in both directions so one can work with another. Turn it on for both machines. When it is off, the machine refuses new links, calls over links, and tasks on another machine.
Default
Off
My suggestion
On for both machines you want to use togetherLinks work only when both machines have it on. Turning it off later lets tasks already running on the other machine finish.
In Settings → Devices, controls whether agents get the device tools. When it is off, the device tools are hidden from agents. It is separate from the Device panel switch.
Default
Off
My suggestion
On when you want agents to verify mobile work on a simulatorTurning on the Device panel alone never grants device control. Agents get device tools only after this is turned on.
An alpha switch in Settings → Labs. When it is on, together with Use devices, you can add a Simlock pool as a device source using an agent token.
Default
Off
My suggestion
Leave off unless you have a Simlock pool and an agent tokenIt is alpha, and adding a Simlock pool needs an agent token. While it is off, the server ignores configured Simlock hosts.
Sets how much of the open thread is read to a screen reader, from turn status only up to more detail. The agent's reasoning text is never read.
Default
Status only
My suggestion
MessagesStatus only says when a turn starts; Messages also reads the agent's in-between messages, which helps you follow progress without looking at the screen. Reasoning text is never read.
A Settings → Chat option that lets a screen reader announce the open thread and what it is doing when you switch to T3 Code.
Default
Off
My suggestion
On for screen reader usersWhen it is on, switching back to T3 Code tells you which thread is open and what it is doing, without navigating into the page.
Typesets LaTeX in messages and Markdown previews. It is under Settings → Appearance.
Default
Off
My suggestion
On for math-heavy workWhen this is off, math parsing and KaTeX stay out of rendering entirely. Turn it on only when your projects use LaTeX, so equations display as typeset math.
Controls whether the main sidebar, right panel, and terminal drawer animate when they open and close.
Default
0 ms (panels open and close immediately)
My suggestion
Up to about 150 ms if you want visible motionA short duration makes panel changes easier to follow. The default of zero avoids per-frame layout work, so keep it short, or leave it at zero on slower machines.
The shortcut that captures the current window. Select it in Settings, press the new keys, then Save.
Default
Not stated in the docs
My suggestion
A modifier chord if Both Shift keys fires accidentallyYou can set a new shortcut in Settings; a chord must include a modifier, which helps avoid accidental captures while typing.