Vizhi for Claude Code

Physical hardware controls for Claude Code on the Logitech MX Creative Keypad. macOS and Windows.

Press a button. Ship code.

Vizhi for Claude Code (formerly Claude Console) turns the MX Creative Keypad's nine LCD keys into a control surface for Claude Code: a key per session, one‑press prompts and git actions, answering its permission prompts, terminal navigation, screenshots into the conversation, and fully offline voice — dictate a prompt or jump into a project by speaking its name. No cloud, no API keys.

---

Features

• A key per session — run Claude in several Terminal tabs and each gets its own key: the project name on a black face, and a bar along the bottom saying what it's doing — Thinking, Waiting, Allow?, Complete. Press one to focus that tab and point every other key at it; the pinned session's bar is highlighted.
• Answer permission prompts — Yes confirms the request Claude is waiting on, No dismisses it, and neither guesses: with nothing to approve they beep and do nothing. Up / Down / Return walk any menu. Details.
• See what needs an answer — both answer keys light amber when Claude wants permission; Yes turns red when approving would run a destructive command (git push, rm -rf, sudo…), while No stays amber. The session's own key reads Allow?. Legend.
• Live status — Cost, context usage and activity read from Claude Code's live-status bridge (opt‑in: the keys are the switch).
• One‑press prompts — Fix Bug, Write Tests, Explore, Explain, Refactor, Review, Optimize, Security, Document, Deploy. One‑word keys, full structured prompts underneath — all customizable, including draft keys you edit before sending.
• Git, through Claude — Commit, Diff, Push, Create PR, Status, Log.
• Screenshot into the conversation — press, drag a region, and the image is handed to the current session with the cursor waiting for your question. Details.
• Offline voice — Dictate sends what you said straight away; Voice Draft leaves it in the input box so you can fix mis‑hearings first; Go to Project opens a project by its spoken name. whisper.cpp transcribes on your machine.
• Terminal & session nav — activate the terminal, new tab, new Claude session, next/prev tab, plus New Claude (Window) and Next/Prev Window if you prefer windows. Terminal.app and iTerm2.
• Model & modes — Model opens the /model picker and keeps a consistent brain icon; Mode cycles normal → auto‑accept edits → plan; plus Compact, Context, Clear, Exit, Tab (accept an autocomplete and run it).
• Types where it should — every key finds Claude's own tab (Terminal.app or iTerm2) and types there, so a press can't land in Slack or a browser because you glanced away. Can't find it? It beeps and types nothing.

See PRIVACY.md — everything runs on your own machine.

Requirements

Both platforms

• Logitech MX Creative Keypad + Logi Options+ 6.4 or newer (installs the Logi Plugin Service). 1.4.0 onwards is built for the .NET 10 runtime that ships with Plugin Service 6.4 — on an older Options+ the plugin will not load, so let Options+ update itself first. (Staying on an older Options+? The last build for it was 1.3.4 — ask via Support.)
• Claude Code CLI, installed natively (WSL sessions are not visible to the plugin — see Windows notes).

macOS

• Apple Silicon (whisper.cpp uses Metal).
• Terminal.app or iTerm2. Every typing key targets Claude's own tab in whichever of the two it runs in; sessions in other terminals are not shown.

Windows

• Windows 10/11, x64. The package carries x64 helpers only; arm64 can be built but is not shipped or tested.
• Windows Terminal for the tab/window navigation keys. Typing works in Windows Terminal, classic conhost and VS Code's integrated terminal — injection addresses the console directly, not the focused window — but the nav keys drive wt.exe, so they need Windows Terminal.
• The keypad layout is imported by hand, not auto-installed — see Windows notes.

To build from source, see docs/development.md.

Windows notes

The Windows build reaches Claude a different way than macOS does, and a few differences are worth knowing before you install.

Typing is if anything safer than on macOS. macOS focuses Claude's Terminal tab and types into it in one atomic AppleScript run. Windows writes key events straight to the target session's console handle, which has no relationship to the focused window at all — a keypress reaches the intended Claude session or nothing whatsoever, and it cannot leak into another application even in principle.

Import the Windows layout. Same as on macOS — the plugin is universal and carries no layout — but the file is Windows-specific, because a profile is bound to an application and the host here is Windows Terminal: Options+ → your keypad → profile menu (⋯) → Import → pick profiles/VizhiClaudeCode-WindowsTerminal.lp5. (You can also just drag Vizhi for Claude Code's actions onto keys yourself; nothing depends on the profile.)

The keypad can't follow your eyes between tabs. Windows Terminal exposes no supported way to ask which tab is in front, so with several idle sessions open the plugin can't tell which one you're looking at. One session needs no pin and just works. So does a single session with a permission prompt up, even when others sit idle at their prompts — a pending approval outranks an idle session — and so does "exactly one session is waiting on you". Beyond that, press a session key first — pinning is exact, and every subsequent key goes to that session until you pin another or press the same key again to release it. Pressing a session key also brings its tab to the front.

When Windows Terminal is not open, terminal-dependent keys refuse safely. They beep and post a "Windows Terminal required" warning in Options+ instead of issuing a wt.exe command that silently does nothing or opens an unrelated window. New Claude (Window) remains available because its job is to create that first Windows Terminal window. Direct typing keys do not depend on wt.exe. Running Claude in a classic Command Prompt or PowerShell console? Set Settings → System → For developers → Terminal to Windows Terminal and start a new session.

Voice works on Windows since 2.2.0 — the package carries a Windows whisper bundle, and the model downloads on first use like on macOS.

Not available on Windows: Next/Previous Window (an OS-level gesture wt.exe cannot express — those keys log and do nothing), and the macOS on‑screen Turn on / Turn off dialog for live status (the second press is the switch). Opening a project always uses a fresh tab rather than reusing an idle one, because Windows Terminal offers no way to tell a busy tab from an idle one and guessing wrong would type into a live session.

Elevated sessions cannot be controlled. Options+ runs unelevated, and Windows blocks the console attach across integrity levels. Run Claude unelevated.

Windows uninstall cleans up live-status wiring in 2.3.1. The plugin removes its own hooks and restores a chained status line before Options+ deletes the package. Reinstall and upgrade preserve your previous live-status setup; turn it off first if you want it to stay off. See uninstall and recovery details.

Install (released plugin)

Install from the Logi Marketplace inside Options+, or download the latest ClaudeConsole_<ver>.lplug4 from vizhi.dev, then:

1. Double-click it — Logi Options+ registers the plugin. (Or, with the Logi Plugin Tool: logiplugintool install ./ClaudeConsole_<ver>.lplug4.) If macOS blocks it, right-click → Open, or run xattr -dr com.apple.quarantine ClaudeConsole_<ver>.lplug4.
2. Put the keys on your terminal. The plugin is universal: it binds to no application and ships no layout of its own, so nothing appears on the keypad until you give it keys. The quickest way is the ready-made layout — see Import the ready-made layout just below (two clicks). Or build your own: in Options+ add Terminal (or iTerm2) as an application, then drag any Vizhi for Claude Code actions onto its profile.
3. On first use, grant Accessibility to the Logi Plugin Service (so it can type into your terminal). For voice, press the Dictate key and grant Microphone when prompted — the helper and speech model install themselves on first use. The Screenshot key asks for Screen Recording the first time.

The live Cost / Context / Activity keys are opt-in: the plugin never edits your Claude Code settings on its own — they read Set up until you press one of them, and the press itself is the prompt. See The live status bridge.

Import the ready-made layout

Rather than mapping nine keys by hand, import the ready-made profile to get the full layout instantly. This is the normal setup, not a fallback — since 2.2.0 the plugin is universal (no application binding, no packaged profile), so the layout is something you import, exactly once.

1. Download VizhiClaudeCode-Terminal.lp5 from the layouts page on vizhi.dev, or take it from profiles/ in this repo. For iTerm2 use VizhiClaudeCode-iTerm2.lp5; on Windows use VizhiClaudeCode-WindowsTerminal.lp5.
2. In Logi Options+ → your MX Creative Keypad, open the profile menu (the ⋯ / profile dropdown) → Import Profile → pick the .lp5.
3. It imports as a profile for that terminal's own entry in Options+ (Terminal or iTerm2), five pages deep, so it activates whenever that app is frontmost. Rearrange or rebind any key afterward. Use both terminals? Import both files.

Page	Keys (top row → bottom row)
1	Session 1 · Session 2 · Session 3 / Clear · No · Yes / Esc · Tab · Dictate
2	Cost · Model · Compact / Up · Return · Down / Scroll Up · Scroll Down
3	Explore · Explain · Review / Optimize · Refactor · Write Tests / Document · Fix Bug · Security
4	Go to Project · New Tab · Next Tab / Prev Tab · New Claude · Exit
5	Commit · Create PR · Diff / Log · Push · Status

Notes:

• Install the plugin first (step 1 above) so the imported keys resolve to real actions.
• Import once. Reinstalling or updating the plugin never touches your profile — it belongs to the terminal's entry in Options+, not to the plugin — so the keys simply light up again after an update.
• The keys type only into Terminal.app and iTerm2: every typing key focuses Claude's own tab in the app that owns it before it types — it will not type into Ghostty, Warp, an editor's terminal or any other app. Sessions running in another terminal are not shown on the Session keys at all (the plugin log names each one it skipped). With sessions in both apps, each key goes to the right one; New Tab, New Claude and Go to Project open in the terminal you are using. In iTerm2, Go to Project reuses an idle tab only with shell integration installed; otherwise it always opens a new tab.
• Session 4–6, Screenshot, Voice Draft, Context, Activity, Mode, Terminal and the window keys exist as actions but aren't on the shipped pages — drag them on wherever you like.
• Already have a Terminal or iTerm2 profile you like? Skip the import and drag the Vizhi for Claude Code actions onto it instead.
• Don't open the live keys in the Options+ icon editor (sessions, Yes/No, Cost/Context/Activity): it freezes them into a snapshot. Why, and the fix.

The live status bridge

The live keys read state files under a private /tmp/claude-console/ directory that Claude Code writes via a status‑line handler (Cost / Context) and five hooks (Activity, and the approval badge). Everything in it is owner‑only (0700 dirs / 0600 files), so your prompts and session state are never readable by other users on the Mac.

Turning this on edits your Claude Code settings, so adding the wiring happens only when you ask — installation never opts you in. The switch is the key itself: press a live key once and it flashes Press again, posts a card in Options+ stating the change and, on macOS, asks on screen — Not now / Turn on. Turn on, or a second press of the same key within 15 seconds, merges a statusLine handler + five hooks into ~/.claude/settings.json. It only appends hooks that aren't already there, chains an existing statusLine (yours still renders), and takes a rolling backup — settings.json.claude-console.bak is rewritten before every change the plugin makes. Everything it does not own is written back exactly as it was found — formatting, quotes, non‑ASCII text, the final newline — so a diff of the file shows only the plugin's own entries. An update may replace only commands already recognisably owned by Vizhi for Claude Code with their current missing-handler-safe form; it never adds wiring or touches user commands. On both platforms the keys come alive with each running session's next activity: the key reads Turned on, and no restart is needed. Whenever live status is off, the Yes / No keys read the same Set up / Off word the live keys do, and every session key's state bar reads Set up / Status off: the approval badge, the answer keys and the state bars all depend on this wiring, and without it nothing they could show would be current.

To take it back out: hold a live key (a long press) and choose Turn off — or hold it again within 15 s — or, without the keypad, bash ~/.claude/claude-console/scripts/uninstall.sh --unwire. Either removes only the plugin's entries and leaves a marker (~/.claude/claude-console/no-autowire) so the keys read Off rather than Set up. Pressing a live key turns it back on.

Without the hooks the Activity key reads Ready and never changes; the Context key needs only the status line, and turns amber at 75% / red at 90% so you compact before an auto‑compaction. To wire it by hand instead, or to see exactly what is written, see docs/live-status-bridge.md.

Several sessions at once

Run Claude in more than one Terminal tab and each session gets its own key in the Sessions group: the project (directory) name on a black face and a state bar along the bottom — Thinking while it works, Waiting when it has stopped for you, Allow? when a permission prompt is up, Complete when the turn is done. Press one to jump to that tab.

Pressing a session key also pins every other key to it, which is the point: you can approve a prompt in session 2 while looking at session 1, or while reading a browser. The pinned session's bar is highlighted; the others are grey. The pin holds until you press another session key, press the same key again to release it, or that session exits — switching Terminal tabs does not move it. With only one session running nothing changes; if you haven't picked a session and exactly one has a permission prompt up, that's the one that gets your Yes — sessions merely idle at their prompt don't compete with it — and failing that, the one session waiting on you at all. When it's genuinely ambiguous the plugin won't guess; it beeps instead of answering the wrong Claude.

Notes:

• A pin aims the keys; it doesn't freeze the readouts. Cost and Context show the session in the tab you're looking at, while Yes/No, Tab and the typing keys act on the pinned one — so you can watch one session while answering another. The amber badge on Yes always describes the session the answer keys will actually answer. (On Windows there's no way to detect the frontmost tab, so both follow the pin.)
• Slots are stable. A session keeps its key for as long as it lives — close one and the others stay put, so you don't approve the wrong session out of muscle memory. The freed key is reused by the next session you start. Six sessions get keys; the shipped layout shows three — drag Session 4–6 onto a page if you run more.
• A new session takes a key immediately (labelled "Claude Code" until it reports its project), and a closed tab clears within about two seconds. Go to Project releases the pin, since it starts a session somewhere new.
• On macOS, Terminal.app or iTerm2, like the rest of the plugin; on Windows, Windows Terminal — see Windows notes.

Using voice

• Dictate — press (you'll hear a Tink), say your prompt, press again. It transcribes locally, types the text into Claude's Terminal tab, and sends it.

• Voice Draft — same flow, but the transcript is only typed, not sent: it sits in Claude's input box so you can fix anything whisper misheard, then submit with Return (keyboard or the keypad's Return key). Use Dictate for quick prompts, Voice Draft for anything long enough to mis-transcribe.

• Go to Project — press, say a project name (e.g. "indie app autopilot"), press again. Opens a new tab in that project running claude; reuses an idle shell tab, or opens a new one if claude is already running.

  Where it looks: projects any session is already open in, plus every git repository within three levels of your home folder — and, once a folder is seen to hold several repositories, its other subfolders too. If your projects live somewhere unusual, list the folders that contain them in ~/.claude/claude-console/project-roots, one per line (~ allowed, # for comments); every subfolder of those becomes matchable and the automatic search is skipped. If nothing matches what you said, the key reads No match and, the first time per plugin load, a card in Options+ names where it looked and the project-roots file.

Start and stop a recording with the same key — each voice key is its own start/stop toggle. Pressing a different voice key while one is recording stops it rather than starting a second recording, and the transcript still goes where the key you started with intended. While a transcript is being produced, a further press is ignored with a beep.

First use prompts once for Microphone permission (granted to the helper, not the daemon). A failed dictation says so on the key, with a beep: Mic denied (allow ClaudeVoiceHelper in System Settings → Privacy & Security → Microphone — or tccutil reset Microphone com.rshankar.claudeconsole.voicehelper and re-grant on the next press; a rebuilt or re-signed helper resets the grant), No speech (it recorded but heard nothing usable), Model loading (the ~142 MB base.en model is still downloading — it's fetched and checksum‑verified on first use, and a card in Options+ says when the download starts, when it is ready, and if it fails; to pre‑seed it, drop ggml-base.en.bin at ~/.claude/claude-console/whisper/), No target (nothing is pinned and no single session is the obvious place to type — press a session key first), Not typed (the session was known but the keystrokes did not land). On Windows the key reads Starting until the helper reports the microphone is live — a second press during that moment cancels — and No helper / No whisper if the voice programs are missing from the install.

Answering Claude's questions

When Claude asks something, answer from the keypad instead of the keyboard:

• Yes / No — answer the permission prompt Claude is waiting on: Yes confirms it (the highlighted option), No dismisses it without running the tool. They act only when the plugin can see a pending approval; with nothing to approve they beep and do nothing rather than guess — typing a word at a hidden menu, or sending a bare Return at an idle prompt, is how the wrong thing gets approved. For plain‑text questions, type into the session. While live status is off they read Set up / Off on a grey tile instead of green and red, and a press posts a card in Options+ saying which key turns it on — without the wiring they cannot see a prompt at all.
• Up / Down / Return — navigate and confirm any selection menu: multiple‑choice questions (AskUserQuestion), plan‑mode confirmation, or a permission prompt whose other options you want.

The approval badge

The keys tell you that an answer is pending and what Yes would approve. When Claude asks for permission, both Yes and No carry a small filled dot in their top-right corner and the session's own key reads Allow?:

Badges	Meaning	What to do
(none on either key)	Nothing is waiting for an answer.	—
🟡 Amber on Yes and No	Waiting on you, and approving is routine — reading a file, running a test, an edit.	Press Yes or No.
🔴 Red on Yes, 🟡 amber on No	Yes would run something destructive or outward-facing; No would reject it.	Look before pressing Yes, or press No to reject.

Red is triggered by the pending command matching one of the patterns in src/Core/RiskClassifier.cs — sudo, rm -rf, git push, git reset --hard, git clean -fd, --force, dd of=, mkfs, chmod 777, drop table, delete from, piping a download into a shell, kubectl delete, terraform apply/destroy, npm publish, gh release create, killall, shutdown…

Two things worth knowing:

• It's a hint, not a gate. Claude Code's own prompt is still what actually holds the command, and the classifier deliberately leans toward warning you unnecessarily — a badge that stayed quiet on a real git push --force would be worse than one that cries wolf. (Bare generic flags like a lone -f are not flagged: a badge that lights on every other command teaches you to ignore it.)
• The badge describes the session the answer keys are pinned to; with several sessions, the Allow? bar shows which one wants attention.

The badge clears itself as soon as the session stops waiting, or the moment you answer from the keypad — a rejection fires no hook, so the plugin clears it on its own. It's powered by a PermissionRequest hook the plugin wires up for you (see The live status bridge); without it Yes/No have nothing to see — they read Set up / Off and a press beeps and explains itself in Options+.

These keys focus Claude's Terminal tab automatically before typing (verified by its TTY), so they work even when another app is frontmost — if Terminal isn't running or the tab is gone, they beep and type nothing.

Screenshot

Screenshot (Core group) captures a region of the screen with the system's own picker (Shift+Cmd+4 style) and hands the image to the current conversation: the file path is typed with a short instruction, without pressing Return, so you add your question and send. First use asks for Screen Recording for the Logi Plugin Service; if a capture later produces nothing, that grant was refused. On Windows the picker is the system snipping overlay and the path is typed into the session directly, like every typing key; only when no session is open does it start one, and that needs Windows Terminal.

Accepting autocomplete, switching modes, scrolling

• Tab — completes Claude Code's highlighted suggestion (a /slash command or an @file mention) and submits it in one press. Because it always presses Return, use Up / Down / Return if you want to complete without sending.
• Mode — sends Shift+Tab, which cycles Claude Code's input modes: normal → auto‑accept edits → plan.
• Scroll Up / Scroll Down — page through the transcript (Page Up / Page Down). In classic mode that scrolls Terminal's scrollback (keep a generous limit in Terminal ▸ Settings ▸ Profiles ▸ Window); in /tui fullscreen mode Claude Code scrolls its own buffer.

All of these focus Claude's Terminal tab before sending keys.

Customizing prompt keys

Every Prompts key is yours to change. The keys show one-word labels (Explore, Review, …), but each sends a full, structured prompt — Review, for instance, asks for a senior-engineer pass with file:line, severity, and a concrete failure scenario per finding; Deploy runs the checks but stops for your approval before anything goes public.

Vizhi for Claude Code stores these keys in ~/.claude/claude-console/prompts.json; Vizhi for Codex uses ~/.codex/vizhi/prompts.json. On its first load, Vizhi copies an existing valid Claude prompt file without changing the original. Future edits are independent. If no source exists, defaults are seeded. A malformed source is left untouched, with defaults used in memory and migration retried on the next load. Edit it to suit how you work — reword any prompt, relabel or re-icon a key, delete keys you never press, or add your own macros. Each entry becomes its own bindable key, and there's no fixed count:

[
  { "id": "ship", "label": "Ship", "icon": "create_pr",
    "prompt": "Run the tests; if green, commit with a conventional message and open a PR." },
  { "id": "standup", "label": "Standup", "icon": "log",
    "prompt": "Summarize what we changed today as 3 standup bullets." },
  { "id": "explain_this", "label": "Explain…", "icon": "explain", "submit": false,
    "prompt": "Explain how this code works, focusing on " }
]

• id — unique key id · label — text under the icon · prompt — typed into the terminal on press.
• icon — an embedded icon basename. Since 2.2.0 every action icon is the same copper monochrome (colour is reserved for state), so the name only picks the glyph: fix_bug, write_tests, explore, explain, refactor, review, optimize, security, document, deploy, commit, diff, push, create_pr, status, log, project, terminal, screenshot, compact (an unknown name falls back to text).
• submit (optional, default true) — set false to make a draft key: it types the prompt but doesn't press Return, so you can edit or finish the sentence before sending it.

Reload the plugin to pick up edits (restart Logi Options+ / killall LogiPluginService). On either product, an empty JSON array ([]) removes every prompt key. Deleting the file restores the built-in defaults in Vizhi for Claude Code, and in Vizhi retries the one-time copy from Claude's file. Your edits are permanent — once you change anything in the file, no plugin update will overwrite it; only an untouched factory prompts.json is upgraded in place when a release improves the defaults.

Key map

Group	Keys
Sessions	Session 1-6 — one key per running Claude session: project name + state bar (press to focus it and pin the other keys to it)
Core	Cost* · Activity* · Context* · Model · Screenshot · Esc · Mode · Tab · Compact · Clear · Exit
Answer	Yes · No · Up · Down · Return
Prompts	Fix Bug · Write Tests · Explore · Explain · Refactor · Review · Optimize · Security · Document · Deploy
Git	Commit · Diff · Push · Create PR · Status · Log
Scroll	Scroll Up · Scroll Down
Terminal	Terminal · New Tab · New Claude · Next Tab · Prev Tab · New Claude (Window) · Next Window · Prev Window · Go to Project (voice)
Universal	Dictate · Voice Draft

* live display, updates from the status line once live status is on.

How it works

MX Creative Keypad → Logi Plugin Service → C# plugin (BridgeManager)
                                                  ↕  file IPC in /tmp/claude-console (owner-only)
Claude Code ← status line (bash) + voice helper (Swift + whisper.cpp)

File‑based IPC under a private /tmp/claude-console/ root (0700 dirs / 0600 files); action keys focus Claude's own tab in Terminal.app or iTerm2 (verified by TTY) and type via osascript; voice records through a notarized helper app that owns its own Microphone permission, then transcribes with a bundled, self‑contained whisper-cli (no Homebrew at runtime). This repo builds one package per agent from a shared engine — Vizhi for Claude Code and Vizhi for Codex for OpenAI's Codex CLI; both can be installed side by side. Architecture in docs/multi-agent-architecture.md; packaging in SUBMISSION.md.

Troubleshooting

The four most common, in full with the rest in docs/troubleshooting.md:

• Nothing appears on the keypad after installing. Expected — the plugin is universal and ships no layout. Import the ready-made layout.
• Keys show only an exclamation mark / plain text right after building from source. A dev .link and an installed .lplug4 are both registered; keep one.
• The session keys are stuck (one session ever shows, the badge never lights, but presses work). A key opened in the Options+ icon editor is frozen into a snapshot — bash scripts/unfreeze-keys.sh --apply.
• Windows: session/navigation keys beep, or Options+ says "Windows Terminal required". Open Windows Terminal and start the agent there; elevated sessions can't be reached.

The plugin's log is at ~/Library/Application Support/Logi/LogiPluginService/Logs/plugin_logs/ClaudeConsole.log (Windows: %LOCALAPPDATA%\Logi\LogiPluginService\Logs\plugin_logs\ClaudeConsole.log).

Uninstall

In Logi Options+, right‑click the Vizhi for Claude Code plugin → Uninstall and delete the imported Vizhi for Claude Code — Terminal (or — iTerm2 / — Windows Terminal) profile; then run the cleanup script the plugin left outside its own package:

bash ~/.claude/claude-console/scripts/uninstall.sh            # confirm, then remove
bash ~/.claude/claude-console/scripts/uninstall.sh --dry-run  # preview only

It unwires the live status hooks surgically, removes ~/.claude/claude-console/ (voice helper, speech model, your prompts.json), the IPC files, the Microphone grant and any dev .link, and asks before deleting. What it touches, why the order matters, and the Windows caveat: docs/uninstall.md.

Windows 2.3.1 removes its settings wiring during Options+ uninstall. User data and downloaded voice files remain for a future reinstall. On macOS, hooks remove their own entries after the plugin folder has been missing for over a minute across two runs; the runtime home stays for the cleanup script above. Failure recovery, upgrade behavior, and removal of saved data are covered in docs/uninstall.md.

Developing

Build, test, package and release notes are in docs/development.md; contributor rules in CLAUDE.md.

Feedback

Found a bug, missing a key you'd use daily, or running a terminal this doesn't support yet? See Support on vizhi.dev — feature requests and "this confused me" reports are equally welcome.

License

Proprietary — see the EULA (EULA.md in this repository) and the privacy policy. Bundled third‑party components (whisper.cpp, the Whisper model) are MIT‑licensed; their licence texts ship with the plugin.