The best way to control Claude Code, Codex and more from any computer or phone

main
md 170 lines 10.1 KB
Rendered Raw
1 # Voice input
2
3 Dictate into the composer instead of typing. The transcript lands in your draft
4 for review, and only sending the message sends anything to an agent. Dictation
5 also works in citation comments, diff and file annotations, pull request
6 comments, replies, and review summaries, and in the desktop app's browser
7 annotation editor.
8
9 On iPhone and Android, voice input transcribes on the device by default. To opt in
10 to a named environment on that phone, open **Settings → Voice → Transcribe on**. See
11 [Voice input on mobile](./composer.md#voice-input-on-mobile).
12
13 ## Choose where audio goes
14
15 On web and desktop, you choose where transcription runs.
16 Nothing is sent until you choose one: open **Settings → Voice** and set
17 **Transcribe with** to the environment that should transcribe for you.
18 On a new device, the microphone or dictation shortcut opens that setting from the composer.
19 Microphone audio recorded on this device is sent to that environment over your
20 authenticated connection, transcribed there, and discarded. The choice is saved
21 on this device only, so every computer and browser you use asks for it once.
22 Set it back to **Off** at any time; a recording in progress is discarded.
23
24 If the chosen environment goes offline, voice input waits for it. T3 Code never
25 substitutes another environment.
26
27 In the desktop app on macOS 26 or later, you can instead choose **Apple Speech**
28 under **On this Mac**. Audio stays on the Mac and macOS transcribes it. The first
29 recording in a language can take longer while macOS downloads its speech model.
30 **Language** set to automatic uses your Mac's language. Silence filtering and
31 filler-word removal are not available; the dictionary applies only when transcript
32 cleanup is on.
33
34 For cloud dictation, choose **OpenAI** or **Grok Voice Transcribe** via the environment
35 you want to use. If it needs a key, enter one and choose **Save and enable**.
36 You can also choose a [secret provider](secrets.md) instead of **Value** and enter a
37 reference. Unlock the vault on that environment before dictating. Anyone with
38 dictation access to the environment can use the key, but only this device opts in.
39 If a key is already saved, just confirm the provider. After
40 you stop, the environment forwards your recording and any vocabulary hints to
41 the selected provider, which bills its API account. No local model download or chat
42 subscription is needed. Switch back to **This device** on mobile or **Off** on
43 web and desktop to stop sharing. Removing the key stops that provider for everyone
44 on the environment.
45
46 **OpenRouter** works the same way but has no default model: pick one from its live
47 speech-to-text catalog when you enable it, or later under **Manage OpenRouter**.
48 The model choice is yours alone and needs no settings access, so anyone with
49 dictation access can pick a different model without touching the saved key.
50 OpenRouter chooses who runs the model. T3 requests zero data retention; your
51 OpenRouter account can also allow safety retention. If no endpoint qualifies,
52 dictation fails instead of switching models.
53 Vocabulary hints are not sent to OpenRouter; dictionary spellings are applied
54 after transcription. If OpenRouter stops listing your model, dictation stops
55 until you choose another.
56
57 ## Set up a model
58
59 In **Settings → Voice → Transcription**, download a model and choose **Use model** if it
60 isn’t already selected. **Test model** lets you try it without sending a message.
61 If you use the composer microphone before a model is installed, it offers to download one.
62 The recommendation uses your Voice language, or your system’s preferred languages with **Auto-detect**.
63
64 If another app already downloaded one of these models, you don’t need a second copy.
65 **Download** first looks in Handy’s model folder and the Hugging Face cache on that
66 environment and uses a copy it finds there instead of downloading. For a copy anywhere
67 else, on web or desktop, paste the path of the model file or its folder under the model
68 list and choose **Use existing**. The path is on the environment
69 that transcribes, not on the device you are using. T3 Code checks that the file is exactly a
70 model from its list and then uses it in place. Removing the model in T3 Code leaves your file
71 alone; if the other app deletes it, download or link the model again. Files in another
72 format, such as Core ML, ONNX, or MLX builds of the same model, can’t be used.
73
74 Select the languages you dictate in to compare models that support all of them.
75 Accuracy shows how often the model got words or characters wrong in each
76 language; speed is the time to transcribe 30 seconds on a reference CPU or GPU,
77 matching the selected environment's transcription backend. Recommendations use
78 the same reference tier for every selected language. Choose **All languages** to
79 browse without a language preference. Filtering does not switch your model.
80
81 **Advanced** controls GPU or CPU use and when an idle model is unloaded. With hardware set to
82 **Auto**, a recording that fails on the GPU is retried once on that same
83 environment's CPU; a GPU you picked yourself reports the error instead. Nothing
84 is ever retried on another environment. Models marked **Live preview** show
85 text while you speak. On web and desktop, turn off **Advanced → Streaming transcription**
86 to transcribe with the same model after stopping instead. This may help when live
87 transcription falls behind. The preference is saved on this device and only applies
88 to models that support native streaming. Other models can process long recordings in chunks at pauses
89 while you record; models with built-in long-form support transcribe when you stop.
90
91 ## Dictate
92
93 Press the microphone in the composer, or `mod+shift+d` (`mod` is Command on
94 macOS and Ctrl elsewhere). Press again, or the check mark, to finish; `Esc`
95 discards the recording and leaves your draft unchanged. Recordings last up to
96 five minutes. Text is inserted at the cursor or replaces the selection you had
97 when recording started, and the editor is read-only while dictation runs.
98 Changes that reach the draft meanwhile, such as a cited message, are kept;
99 the text then goes after your selection instead of replacing it.
100
101 Pressing Send while dictating finishes the recording and sends the draft once
102 the transcript has been inserted. Cancelling also cancels that queued send. In
103 the other editors, Save, Comment, Reply, Submit review, or Attach works the
104 same way, and the shortcut targets the editor that has focus, or the composer
105 when none does. In the desktop browser's annotation editor, the microphone and
106 the shortcut work while the page has focus; the page you are annotating never
107 sees the recording, which runs in T3 Code itself.
108
109 **Dictation shortcut** in Voice settings picks how the keybinding behaves:
110 **Auto** tap-toggles or records while held, **Hold** records only while the keys
111 are down, and **Toggle** always needs a second press. Change the chord itself in
112 [Keybindings settings](./keybindings.md).
113
114 ## Improve recognition
115
116 - **Language** picks what you speak. **Auto** lets a multilingual model detect
117 it. Models that translate offer **Translate to English**.
118 - **Dictionary** holds names and terms the model should get right. Add an alias
119 under a term when transcription keeps producing a particular misspelling.
120 Your dictionary is saved on this device. The name of the project you dictate
121 in is sent along as a hint.
122 - **Project dictionary** is vocabulary saved with a project on its environment,
123 so everyone who dictates in that project gets the same spellings. Pick the
124 project at the top of Settings to edit it; with an environment selected, the
125 entries apply to every project there. Project spellings win when they overlap
126 your personal dictionary. Each recording carries up to 100 entries from the
127 two combined.
128 - **Silence filtering** is on by default. Your chosen environment skips non-speech
129 audio before transcription. Turn it off to keep those sections. Filtering happens
130 after upload: it does not grant permission to send audio or reduce how much is sent.
131 - **Remove filler words** drops hesitation sounds from finished transcripts.
132
133 Language and these preferences travel with each recording to the chosen
134 environment; nothing else about the project is sent.
135
136 ## Clean up transcripts
137
138 **Transcript cleanup** is a separate choice and is off by default. When on, the
139 finished transcript and up to 1,024 characters each from before, inside, and
140 after your selection are sent to the cleanup model on the draft's environment,
141 which fixes punctuation, numbers, and spoken corrections such as "make it 42,
142 sorry, 24". Depending on the model, that can involve an external service. Audio is never
143 sent to it. Use the built-in instructions or write your own, press **Skip**
144 while cleanup runs to insert the raw transcript, and if cleanup fails the raw
145 transcript is inserted.
146
147 Cleanup uses the environment's text generation model unless you pick a
148 **Cleanup model** for that environment, for example a cheaper or faster one.
149 It can be a text generator instead of a coding agent; see
150 [text generation models](./text-generation.md).
151
152 If cleanup misses the way you correct yourself, set a **Correction cue**: the
153 word or phrase you say to flag a correction, such as "scratch that". Cleanup
154 applies the correction only where the context shows one and drops the cue, and
155 the model hears the cue as a dictionary hint.
156
157 ## Test and troubleshoot
158
159 In **Settings → Voice**, voice input uses the first connected microphone in the
160 **Microphone** list. A microphone joins the list below System default the first
161 time it is seen, so nothing changes until you move it up. If a preferred
162 microphone is unplugged, the next connected one is used.
163
164 **Test microphone** records from that microphone for up to ten seconds and
165 plays it back in your browser; the audio stays on this device and is never transcribed, so silent playback
166 points at the microphone. **Test model** on the selected model shows a transcript using the
167 same environment, language, and dictionary as the composer, but skips transcript
168 cleanup even when it is on, so you see what the model heard. Browser recording
169 needs HTTPS or localhost and microphone permission; for a remote server over
170 plain HTTP, use the desktop app.