Skip to content

7.4 billion tokens per month. 34 free LLM providers. 635 free model endpoints. All behind one /v1 endpoint, plus any custom OpenAI-compatible endpoint. Smart routing, automatic failover, encrypted keys. Personal experimentation only.

TypeScript

fix(media): serve transcription usage at /api/media/usage + show it on the STT tab (#991)

tashdroidb6955f676f4c
diff --git a/client/src/components/media-models.tsx b/client/src/components/media-models.tsx
index ef30b64c..4d992bd6 100644
--- a/client/src/components/media-models.tsx
+++ b/client/src/components/media-models.tsx
@@ -25,7 +25,7 @@ export interface MediaModel {
interface MediaData { models: MediaModel[] }
interface MediaUsage {
- modality: 'image' | 'audio'
+ modality: 'image' | 'audio' | 'transcription'
models: {
id: number
platform: string
@@ -152,6 +152,15 @@ export function MediaModelsView({ modality }: { modality: 'image' | 'audio' }) {
refetchInterval: 30_000,
})
+ // The Audio tab carries both directions of the modality, so the STT section
+ // gets its own usage summary (the endpoint accepts modality=transcription).
+ const { data: sttUsage } = useQuery<MediaUsage>({
+ queryKey: ['media', 'usage', 'transcription'],
+ queryFn: () => apiFetch('/api/media/usage?modality=transcription'),
+ enabled: modality === 'audio',
+ refetchInterval: 30_000,
+ })
+
const toggle = useMutation({
mutationFn: ({ id, enabled }: { id: number; enabled: boolean }) =>
apiFetch(`/api/media/${id}`, { method: 'PUT', body: JSON.stringify({ enabled }) }),
@@ -232,6 +241,21 @@ export function MediaModelsView({ modality }: { modality: 'image' | 'audio' }) {
{t('models.sttHint')} <code className="rounded bg-muted px-1 py-0.5 font-mono text-[11px]">/v1/audio/transcriptions</code>
</p>
+ {sttUsage && sttUsage.models.length > 0 && (
+ <UsageSummaryCard
+ rows={sttUsage.models.map(m => ({
+ label: m.displayName,
+ platform: m.platform,
+ quotaLabel: [m.platform, m.quotaLabel].filter(Boolean).join(' · '),
+ amount: m.requestsMonth,
+ requestsToday: m.requestsToday,
+ }))}
+ total={sttUsage.totalRequestsMonth}
+ requestsToday={sttUsage.totalRequestsToday}
+ unit="requests"
+ />
+ )}
+
{isLoading ? (
<CardSkeleton />
) : sttGroups.length === 0 ? (
diff --git a/server/src/__tests__/routes/media-usage.test.ts b/server/src/__tests__/routes/media-usage.test.ts
index 2f7fcefe..769d3693 100644
--- a/server/src/__tests__/routes/media-usage.test.ts
+++ b/server/src/__tests__/routes/media-usage.test.ts
@@ -34,6 +34,10 @@ function seedModels() {
insert.run('groq', 'playai-tts', 'PlayAI TTS', 'audio', 1, 1, 'WAV/PCM - 30+ voices');
// Disabled rows stay out of the summary entirely.
insert.run('cloudflare', 'melotts', 'MeloTTS', 'audio', 2, 0, 'Shared 10k neurons/day');
+ // STT models live in the same table with modality='transcription'; the
+ // Audio tab lists them under the TTS section, so the usage summary has to
+ // accept the modality too (it used to 400).
+ insert.run('groq', 'whisper-large-v3', 'Whisper Large v3', 'transcription', 1, 1, 'MP3 output - multilingual');
}
function seedRequest(platform: string, modelId: string, type: string, when: string) {
@@ -102,6 +106,27 @@ describe('GET /api/media/usage', () => {
expect(flux.requestsMonth).toBe(0);
});
+ it('reports transcription usage from the same tagged request log', async () => {
+ seedRequest('groq', 'whisper-large-v3', 'transcription', "datetime('now')");
+ seedRequest('groq', 'whisper-large-v3', 'transcription', "datetime('now', 'start of month')");
+ // Same provider AND model id, but TTS traffic: the join must exclude it by
+ // request_type, not just by platform + model_id.
+ seedRequest('groq', 'whisper-large-v3', 'audio', "datetime('now')");
+
+ const { status, body } = await get(app, '/api/media/usage?modality=transcription');
+ expect(status).toBe(200);
+ expect(body.modality).toBe('transcription');
+ expect(body.models.map((m: any) => m.modelId)).toEqual(['whisper-large-v3']);
+
+ const whisper = body.models[0];
+ // Exactly the two transcription requests — the audio one is filtered out
+ // by request_type even though it shares platform and model id.
+ expect(whisper.requestsMonth).toBe(2);
+ expect(whisper.requestsToday).toBeGreaterThanOrEqual(1);
+ expect(whisper.quotaLabel).toBe('MP3 output - multilingual');
+ expect(body.totalRequestsMonth).toBe(2);
+ });
+
it('reports no budget field — media quota labels are not token budgets', async () => {
const { body } = await get(app, '/api/media/usage?modality=audio');
for (const m of body.models) {
diff --git a/server/src/routes/media.ts b/server/src/routes/media.ts
index 4101e116..a04a378c 100644
--- a/server/src/routes/media.ts
+++ b/server/src/routes/media.ts
@@ -48,11 +48,14 @@ mediaRouter.get('/', (_req: Request, res: Response) => {
// token counts are reported. As with embeddings there is no budget
// denominator — `media_models` only carries a free-text `quota_label`
// ("Shared 10k neurons/day", "MP3 output - multilingual", which is not even a
-// quota) — so the summary shows spend and the label verbatim.
+// quota) — so the summary shows spend and the label verbatim. Transcription
+// rows are logged the same way (request_type='transcription', see
+// logMedia), so the Audio tab's STT section can show the same per-model
+// counts instead of dead-ending on a 400.
mediaRouter.get('/usage', (req: Request, res: Response) => {
- const parsed = z.enum(['image', 'audio']).safeParse(req.query.modality);
+ const parsed = z.enum(['image', 'audio', 'transcription']).safeParse(req.query.modality);
if (!parsed.success) {
- res.status(400).json({ error: { message: 'modality must be image or audio' } });
+ res.status(400).json({ error: { message: 'modality must be image, audio or transcription' } });
return;
}
const modality = parsed.data;