Node-based terminal manager for AI coding agents — tmux-backed terminals and parallel agent sessions as draggable nodes on an infinite pan/zoom canvas. macOS, Linux, and a browser Server Edition.
TypeScript adhdagent-orchestrationai-agentscanvasclaude-codecodexcoding-agentsdeveloper-toolselectrongeminilinuxmacosmultiplexeropencodeparallel-agentsterminalterminal-emulatorterminal-multiplexertmuxworkspace-manager
import { describe, expect, it } from 'vitest'
import {
WHISPER_MODELS,
whisperModel,
WHISPER_DOWNLOAD_BASE,
hasSpeechModel,
modelAfterDownload,
modelAfterDelete,
SPEECH_MODEL_NONE,
type ModelDownloadState,
SPEECH_LANGUAGES,
SPEECH_LANGUAGE_AUTO,
speechLanguage,
speechLanguageLabel,
filterSpeechLanguages,
modelKnowsV3Languages
} from './speech'
import { DEFAULT_SETTINGS } from './types'
describe('whisper model catalog', () => {
it('matches the spec table exactly', () => {
expect(WHISPER_MODELS.map((m) => m.id)).toEqual(['tiny', 'base', 'small', 'large-v3-turbo'])
expect(WHISPER_MODELS.map((m) => m.file)).toEqual([
'ggml-tiny.bin', 'ggml-base.bin', 'ggml-small.bin', 'ggml-large-v3-turbo.bin',
])
// Only tiny is free — the Pro split the spec mandates.
expect(WHISPER_MODELS.filter((m) => !m.pro).map((m) => m.id)).toEqual(['tiny'])
})
it('looks up by id and rejects unknowns', () => {
expect(whisperModel('base')?.approxMB).toBe(142)
expect(whisperModel('nope')).toBeUndefined()
})
it('download base is the whisper.cpp HF repo', () => {
expect(WHISPER_DOWNLOAD_BASE).toBe('https://huggingface.co/ggerganov/whisper.cpp/resolve/main/')
})
it('speech settings default to dictation OFF — nothing downloads until the user asks (#143)', () => {
expect(DEFAULT_SETTINGS.speech).toEqual({
engine: 'whisper',
model: SPEECH_MODEL_NONE,
language: 'auto',
shortcut: 'Cmd+Alt'
})
})
it('hasSpeechModel reads None as off and any id as on', () => {
expect(hasSpeechModel(SPEECH_MODEL_NONE)).toBe(false)
expect(hasSpeechModel('tiny')).toBe(true)
})
})
const list = (...ids: Array<[string, boolean]>): ModelDownloadState[] =>
ids.map(([id, downloaded]) => ({ id, downloaded }))
describe('modelAfterDownload', () => {
it('adopts the first download — the default selection points at a model nobody fetched', () => {
// `tiny` is selected out of the box (see the default above), so a user whose first download is
// `base` was left pointed at a file that is not on disk, and dictation failed.
expect(modelAfterDownload(list(['tiny', false], ['base', true]), 'tiny', 'base')).toBe('base')
})
it('leaves a working selection alone when a second model is tried out', () => {
expect(modelAfterDownload(list(['tiny', true], ['small', true]), 'tiny', 'small')).toBeNull()
})
it('is a no-op when the selection is what was just downloaded', () => {
expect(modelAfterDownload(list(['tiny', true]), 'tiny', 'tiny')).toBeNull()
})
it('adopts when the selected id is not in the catalogue at all', () => {
// A removed/renamed model in an old settings file has nothing behind it either.
expect(modelAfterDownload(list(['base', true]), 'ancient', 'base')).toBe('base')
})
it('adopts out of the None state — starting a download IS asking for dictation (#143)', () => {
expect(modelAfterDownload(list(['base', true]), SPEECH_MODEL_NONE, 'base')).toBe('base')
})
})
describe('modelAfterDelete', () => {
it('falls back to another downloaded model when the selected one is deleted', () => {
expect(modelAfterDelete(list(['tiny', false], ['small', true]), 'tiny')).toBe('small')
})
it('keeps the selection when it is still on disk', () => {
expect(modelAfterDelete(list(['tiny', true], ['small', true]), 'tiny')).toBeNull()
})
it('answers null when nothing is downloaded — there is nothing truthful to select', () => {
expect(modelAfterDelete(list(['tiny', false], ['small', false]), 'tiny')).toBeNull()
})
it('NEVER heals an explicit None — deleting a leftover model keeps dictation off (#143)', () => {
// With None selected and another model still on disk, the old fallback would have re-adopted
// it behind the user's back. An unset state is a legitimate value, not a dangling pointer.
expect(modelAfterDelete(list(['tiny', true], ['small', true]), SPEECH_MODEL_NONE)).toBeNull()
})
})
describe('speech language catalog', () => {
it('covers whisper’s own language set, Polish and the rest included', () => {
// The 7-entry hand-written list this replaced (issue #586) is the floor, not the shape:
// every code whisper accepts must be reachable from the picker.
const codes = SPEECH_LANGUAGES.map((l) => l.code)
expect(codes).toHaveLength(100)
expect(new Set(codes).size).toBe(100)
for (const c of ['pl', 'it', 'pt', 'ru', 'uk', 'nl', 'zh', 'ko', 'hi', 'cs']) {
expect(codes).toContain(c)
}
// The codes the old list had must all still be there — nobody loses their setting.
for (const c of ['en', 'tr', 'de', 'fr', 'es', 'ja']) expect(codes).toContain(c)
// `auto` is NOT a whisper language: it means "no hint", and lives outside the table.
expect(codes).not.toContain(SPEECH_LANGUAGE_AUTO)
})
it('is ordered by English name and fully populated', () => {
const labels = SPEECH_LANGUAGES.map((l) => l.label)
expect(labels).toEqual([...labels].sort((a, b) => a.localeCompare(b, 'en')))
for (const l of SPEECH_LANGUAGES) {
expect(l.code).toMatch(/^[a-z]{2,3}$/)
expect(l.label.length).toBeGreaterThan(0)
expect(l.endonym.length).toBeGreaterThan(0)
// An alias that only repeats the label costs a scan and teaches nothing.
for (const a of l.aliases ?? []) expect(a.toLowerCase()).not.toBe(l.label.toLowerCase())
}
})
it('marks Cantonese as the one large-v3-only entry', () => {
expect(SPEECH_LANGUAGES.filter((l) => l.sinceV3).map((l) => l.code)).toEqual(['yue'])
expect(modelKnowsV3Languages('large-v3-turbo')).toBe(true)
expect(modelKnowsV3Languages('tiny')).toBe(false)
expect(modelKnowsV3Languages('small')).toBe(false)
})
it('names the stored value, and hands back an unknown code instead of eating it', () => {
expect(speechLanguageLabel('auto')).toBe('Auto-detect')
expect(speechLanguageLabel('pl')).toBe('Polish')
// The old <select> rendered a stored code it did not know as blank and overwrote it on the
// next click. A code we cannot name is still the user's setting.
expect(speechLanguageLabel('xx')).toBe('xx')
expect(speechLanguageLabel('')).toBe('')
expect(speechLanguage('pl')).toMatchObject({ label: 'Polish', endonym: 'polski' })
expect(speechLanguage('xx')).toBeUndefined()
})
it('finds a language by code, English name, endonym or alias', () => {
const first = (q: string): string | undefined => filterSpeechLanguages(q)[0]?.code
expect(first('pl')).toBe('pl')
expect(first('pol')).toBe('pl')
expect(first('polski')).toBe('pl')
expect(first('Polish')).toBe('pl')
expect(first('deutsch')).toBe('de')
expect(first('mandarin')).toBe('zh') // whisper's own alias table
expect(first('bengali')).toBe('bn') // CLDR calls it Bangla
expect(filterSpeechLanguages('zzzz')).toEqual([])
expect(filterSpeechLanguages(' ')).toEqual(SPEECH_LANGUAGES)
})
it('folds diacritics, so a speaker can type their own name without them', () => {
const first = (q: string): string | undefined => filterSpeechLanguages(q)[0]?.code
expect(first('turkce')).toBe('tr') // Türkçe
expect(first('cestina')).toBe('cs') // čeština
expect(first('islenska')).toBe('is') // íslenska
// Non-Latin scripts are searched as typed.
expect(first('русский')).toBe('ru')
expect(first('日本語')).toBe('ja')
})
it('ranks an exact code above a name that merely contains it', () => {
// "pl" is a substring of Nepali's endonym neighbourhood; the code must still win.
expect(filterSpeechLanguages('pl')[0].code).toBe('pl')
// Prefix beats substring: "ka" is Georgian's code, and also inside "Slovak"/"Kazakh".
expect(filterSpeechLanguages('ka')[0].code).toBe('ka')
})
})