Audio and video catalog support
This release includes the adapters for the November 1 catalog update: 13 tested media additions, with three duplicate listings disabled (net +10 enabled entries).
- NovAI: daily-free CogVideoX Flash video generation, with asynchronous polling, cancellation, safe media downloads and normal routing/failover.
- Free.ai: MeloTTS, Chatterbox, Qwen3 TTS and VibeVoice speech; Whisper uploaded-file transcription.
- OVH: keyless English, German, Spanish and Italian speech, plus Whisper transcription support.
- Mistral: Voxtral Mini TTS 3.0 speech.
- Google: Gemini 3.5 Transcribe through the Interactions API.
- ElectronHub: support for the catalog's GPT-4o Mini TTS route through the existing speech adapter.
- Cohere: transcription routing and correct classification of its existing speech-recognition entry.
Fixes
- Correct Cloudflare Aura speech payloads and binary audio handling, MeloTTS WAV labeling, and Nova 3 transcription parsing.
- Keep dedicated transcription routes enabled when their legacy duplicate audio listings are disabled. Local user disables remain respected.
Rollout
The signed Oracle catalog is already published as 2026.11.01.fbb00f16: 1,221 total entries, 737 enabled, 57 active providers. Premium receives new catalog entries immediately; Free receives them through the existing 30-day promotion. This release does not bypass that gate or seed new models locally.
Update the app, configure the relevant provider keys, and sync the catalog. Provider allowances are shared and upstream availability/rate limits still apply. NovAI's free output is watermarked.
Verification
All 13 additions passed live application-adapter tests. The adapter PR passed Node 20/22/24 CI and Linux amd64/arm64 Docker builds. Local regression checks passed 4,811 server tests, 178 CLI tests and 436 client tests, plus build, lint, migration and localization checks. Oracle signatures and the unchanged day-0/day-29/day-30 promotion behavior were verified.
Adapter changes: #1797