Repository navigation
Dictation in the terminal, from prism-term-core (#162) - #163
Merged
Merged
Conversation
…, transcribed on this PC (#162) Hold Right Alt over a showing terminal, speak, release: the words are pasted at the cursor and never sent. The feature is the core's, the same one Prism Terminal has; this is Prism's wiring: the pin, the bundled engine, the microphone permission (audio only), a Dictation page of its own in Settings, and the e2e. With no terminal showing the key does nothing. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Owner
Author
|
Installed 0.54.0 on the owner's machine (Prism was not running): exe timestamp 18:14:32 -> 20:37:48, engine folder present (18 files). |
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Found by running it on a GitHub runner: the real-Claude-CLI checks are skipped, loudly, where no claude is installed; and the new-tab check asserts the folder the profile is seeded with, not /Users/, which only ever passed because the owner's repo lives under C:\Users. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…ve itself (#164) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Owner
Author
|
Merging at the owner's word (2026-09-19), pinned to |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #162. Closes #164. Prism's half of PrismTerminal#13 (its PR: Maxaubert/PrismTerminal#22). Owner, 2026-09-19: "make sure this feature is synced and part of the core and should be the same in normal Prism, with dictation as its own tab there too."
What a user gets
Settings > Dictation (its own page under Behaviour). Switch it on, download a model, then hold Right Alt over a terminal, speak, let go: the words are pasted at the shell's cursor, never sent. A pill over the terminal shows a live level meter and the text as you speak; the tab wears a mic mark. Transcribed on this PC by whisper.cpp; audio never leaves the machine. Off by default.
It is the SAME feature as Prism Terminal's, because it is the same code:
prism-term-core. The models and the NVIDIA engine live in%LOCALAPPDATA%\PrismDictation, shared by both apps, so a model downloaded in one is installed in the other. Settings VALUES stay per app.What is Prism's own
npm run fetch:whisperruns the core's own script intovendor/whisper, packaged asresources/bin/whisper(11 MB, with Microsoft's C++ runtime app-local and the MIT notice).mediafor audio only, to Prism's own windows; the camera is refused. Everything else keeps Electron's default, deliberately.TermDock, the mic mark inTabStrip; both read the dictation store themselves, so speaking re-renders neither App nor the strip.Pin
prism-term-core#core-v0.2.0-rc.6. It becomescore-v0.2.0once PrismTerminal#22 merges; merge that one first, then I repoint this and re-run the gate.The terminal gate now runs in CI (#164, rides in this PR)
Owner, 2026-09-19: "we need automated tests to confirm it never conflicts. that's important."
npm run e2e:terminal, the required gate for any core pin bump, ran only on the owner's machine. New workflowterminal-gate.ymlruns it on a GitHub Windows runner, on any PR that moves the pin or touches the terminal's wiring. Measured: green 4 runs of 4, 173 checks, 4 to 6 minutes, the real dictation scenario included. It is what lets PrismTerminal'score-releaseworkflow open a bump PR that proves itself (PrismTerminal#24).Running it on a runner for the first time found three real things:
terminalscenario waited sixty seconds for a realclaudeCLI a runner does not have (now skipped, loudly, where absent);tabsscenario asserted/Users/and called it "the user folder", which only ever passed because the owner's repo lives under the Users folder (it now asserts the folder the profile is actually seeded with).Verification
npm run e2e:terminal(the required gate for a pin bump): 11 scenarios green, now includingdictationanddictationPage.dictationis REAL end to end in Prism: Chromium's fake microphone plays a known clip, the real bundled engine hears it with the Tiny model, and "ask not what your country can do for you" must land on the prompt line with no new prompt after it.npm run e2e: 775 checks pass;rowPaste(real clipboard, flaky on main too) timed out once in the full run and passed alone.🤖 Generated with Claude Code