Managing streams
Last updated: 6 September 2026
What it does
A "stream" is a single terminal pane running a shell or command. You can add, duplicate, and remove streams, and save frequently used configurations as reusable profiles.
How to use it
Add a stream
- Click + Add Stream in the tab bar, or press
Ctrl+N. - Choose from detected Windows Terminal profiles, your saved stream profiles, or create a new one. Type to filter the list.
- Your default shell (set in Settings) appears first.
Right-click menus
- Right-click a stream tab — Copy Working Folder, Open Working Folder, Duplicate, Remove, and the rest.
- Right-click the tab bar — quick-add a new stream from the profile menu.
Grab the working folder
- Click the path in the pane header to copy it. The header flashes Copied.
- Ctrl+click the path (or Open Working Folder on the tab) opens it in Explorer.
- Ctrl+click the stream tab copies the folder without opening the menu. Hover the tab to see the path.
Save a stream profile
- Open Settings > Streams.
- Click + Create New Stream.
- Set the name, command, working directory, accent color, and recording mode.
- Launch saved profiles from the add-stream menu or the Welcome screen.
StreamView seeds default profiles on first run — PowerShell, PowerShell (Admin), Command Prompt, Command Prompt (Admin), Git Bash, Claude Code, Cursor Agent, and Grok CLI. Agent commands do not skip permission prompts; that's a checkbox in the stream editor. Existing installs that already have templates get Cursor and Grok added by name if they're missing.
A workspace pane is a copy of the template at the moment you add it. Changing the saved template later (Claude → Grok, for example) does not rewrite panes already in a workspace. On launch, StreamView copies the template command onto same-named streams that still have the old claude --dangerously-skip-permissions seed — the folder and any environment on the pane stay. Settings > Streams > Apply templates does the same for any pane whose agent no longer matches its template, after showing what will change. Live panes restart so the new CLI actually launches.
Environment variables
Attach a named environment pack (Settings → Environment) when several panes should share the same secrets — for example, two panes on one project attach my-api; other streams do not. Extra variables takes one KEY=VALUE per line on top of that pack. Blank lines and # comments are ignored. Packs and extras are encrypted on disk for this Windows user (DPAPI) — see Environment variables and secrets.
This replaces chaining them through a shell. Instead of cmd.exe /k set AWS_PROFILE=prod && claude, put AWS_PROFILE=prod in extras (or in a pack) and leave the command as claude. Saving on a running stream restarts it when the pack or extras changed, because a running terminal's environment is fixed when the process is created.
Each start re-reads Machine and User environment from Windows, then the pack, then extras. Restart the pane, not StreamView, after a User-env or pack change.
Simplify a long command
Commands tend to grow into a chain that does several jobs at once:
cmd.exe /k cd /d "C:\repos\my-app" && set AWS_PROFILE=dev && claude --dangerously-skip-permissions
That works, but the directory is invisible to everything that wants to know a stream's working directory, the variable can only be set by chaining through a shell, and the real command is buried at the end where the agent dropdown can't recognise it.
When StreamView spots a command like that, a Simplify button appears beside the Command field. It shows exactly what it will move, then splits it into the fields that already exist:
| Field | Becomes |
|---|---|
| Command | claude --dangerously-skip-permissions |
| Working Directory | C:\repos\my-app |
| Environment Variables | AWS_PROFILE=dev |
The stream runs identically afterwards — and because the command is now a bare agent invocation, the Shell Type dropdown recognises it and its option checkboxes light up.
Doing them all at once: if several saved streams share the pattern — which they usually do, since it's a habit rather than a one-off — Settings > Streams > Simplify all handles the lot. It shows how many streams change and what each command becomes, then applies them together. Streams it doesn't understand are skipped rather than guessed at, and already-running panes keep their current settings until restarted.
It's deliberately cautious: it only lifts cd and set segments from the front of a chain, leaves anything it doesn't recognise exactly where it is, and tells you before replacing a working directory you'd already set. A set that appears after a real command is left alone, since that ordering may well be deliberate.
A related leftover is repaired automatically, without the Simplify button: an old launch rewrite turned cmd.exe /k cd /d "path" && claude into claude.exe /d "path" && claude, which made Claude print Unknown command: /d. Restore, Restart, and new launches strip that /d and keep the folder in Working Directory.
AI coding agents
The Shell Type dropdown includes Claude Code, Cursor Agent, and Grok CLI alongside the shells. Picking one fills in the command and shows that tool's options as checkboxes, so you don't have to remember flag names:
| Agent | Options offered |
|---|---|
| Claude Code | Skip permission prompts, Continue last conversation |
| Cursor Agent | Skip permission prompts, Trust this workspace, Read-only (review mode), Continue last conversation |
| Grok CLI | Skip permission prompts, Plan mode, Native scrollback, Continue last session |
| Codex CLI | — |
| Gemini CLI | — |
| Aider | — |
| GitHub Copilot CLI | — |
| OpenCode | — |
Picking any of them sets the command, a sensible accent colour, and a recording mode suited to an agent. StreamView also tells you if the tool isn't on your PATH before a pane opens. Cursor Agent often lives at %LOCALAPPDATA%\cursor-agent\agent.cmd rather than PATH; StreamView uses that path when it finds it.
Welcome has buttons for Claude, Cursor, and Grok. Starting one with no working directory asks which folder. For Cursor, StreamView then adds --trust — you already chose the folder, so the "do you trust this directory?" prompt is redundant. It does not turn on skip-permissions (--force / --dangerously-skip-permissions / --always-approve); that stays a warned checkbox, off by default.
Claude Code, Cursor Agent, and Grok CLI currently offer option checkboxes, and every flag in them was read from that tool's own help output rather than guessed. Codex, Gemini, Aider, Copilot CLI, and OpenCode are recognised and pre-filled but have no options until their flags have been verified; add any you need in the command box.
Skip permission prompts asks for confirmation before it turns on. It removes the agent's confirmation step entirely, so it will change files and run commands in the working directory without asking. That's the intended behaviour for a trusted sandbox and a poor idea anywhere else — especially combined with Run as administrator.
Read-only (review mode) is the one worth knowing about for multi-agent work: a Cursor pane in --mode ask can analyse and report but cannot edit, which makes it safe to run in the same folder as an agent that is writing. That's the "Claude codes, Cursor reviews" setup without a second checkout.
The checkboxes and the command box stay in step both ways — edit the command by hand and the boxes follow.
Restarting a stream
The ↻ button in a running pane's header stops the process and starts it again with the same command, working directory, and settings. It appears next to ■ while something is running.
Mainly for the case where a tool tells you an update is installed and asks you to restart — including Claude Code and Cursor Agent updating themselves in a pane.
A restart is a clean slate: the pane's scrollback is cleared along with the old process. Session history already written to disk is kept — the old run's history is closed properly and the new run opens its own. For an AI session it starts a fresh conversation, but both Claude Code and Cursor store their transcripts themselves, so claude --continue or cursor-agent --continue picks up where you were.
Run as administrator
Tick Run as administrator under the command to make a stream open elevated. Starting it then shows a Windows UAC prompt — every time, by design; there's no way to remember the answer.
The checkbox is a view over the command itself: ticking it puts sudo (or gsudo) in front of what you typed, so the saved profile stays a plain command line and the built-in Admin templates already show it ticked. Untick to get your original command back.
Why a Windows setting is involved
There is no built-in command that elevates a terminal you're already sitting in. runas switches user rather than elevating, and Start-Process -Verb RunAs opens a new window — no use inside a pane. Raising a running console in place needs a broker: Windows' own sudo in Inline mode, or gsudo.
StreamView can't be that broker itself. A terminal pane's process is created with CreateProcess, which has no route to a consent prompt, so the app is structurally unable to raise a UAC dialog for a pane no matter what it does.
What it can do is make the one-time setup a single click. When the checkbox reports that sudo is off, Turn on sudo for me enables Inline mode for you — Windows asks you to approve the change once, and admin streams work from then on. Open Windows Settings takes you to the same toggle to do it by hand. (Note that sudo config can't be used here: sudo refuses to run at all while it's disabled, so the setting has to be written directly.)
What the pre-flight tells you
Underneath the checkbox StreamView tells you whether elevation will actually work on this machine before you launch, because it can fail several different ways that all look identical afterwards — the pane just closes instantly:
- Windows sudo turned off — enable it at Settings > System > For developers > Enable sudo, set to "Inline".
- sudo set to "In a new window" — the admin shell opens in a separate window and the pane closes. Inline is the only mode that works in a pane.
- sudo set to "With input disabled" — the shell starts but can't receive keystrokes.
- No sudo at all (needs Windows 11 24H2 or later) — install gsudo with
winget install gsudo.
If StreamView is itself running as administrator, every pane is already elevated and the checkbox is shown as settled rather than offered.
Worth knowing before you elevate an AI agent. Running an agent as administrator and skipping its permission prompts removes both safety layers at once — it can then change anything on the machine without asking. There's also a practical catch: an elevated agent writes admin-owned files into
%USERPROFILE%\.claude(and into your repos), which your normal non-elevated panes then can't modify. Most work needs admin for one command, not a whole session — runningsudo <that command>inside an ordinary pane is a far smaller blast radius.
Troubleshooting
- A stream closed immediately after starting — the command may be wrong, or it may require elevation. See Troubleshooting, especially "Admin / elevated terminals."
- My Windows Terminal profile isn't listed — StreamView reads Windows Terminal profiles at startup; restart the app after adding a profile there.
Related
Still stuck? In the app, use Help → Contact support, or open a ticket here.