cartesia
Package
Section titled “Package”@ai-sdk/cartesia
Factories
Section titled “Factories”AIProviders.cartesia
Capabilities
Section titled “Capabilities”- Transcription
- Speech
Known ink-whisper and dated ink-whisper-* snapshots use REST batch
transcription. Every other current or future ID uses the existing streaming
protocol; select streamingTranscriptionModel(...) for those models.
Cartesia REST speech, batch transcription, and Ink 2 streaming transcription mirror upstream 3.0.63: CARTESIA_API_KEY, Bearer auth, Cartesia-Version default 2026-03-01, custom headers and Swift base-URL override, versioned user-agent suffix, speech/transcription aliases, /tts/bytes JSON speech, /stt multipart batch transcription, and /access-token plus duplex /stt/turns/websocket or /stt/websocket streaming. Streaming supports raw PCM/G.711 encodings, server turn detection or explicit finalization, partial/final events, raw chunks, token-redacted metadata, cancellation, and an injectable AIDuplexWebSocketTransport. Full realtime speech-session generation remains a broader protocol gap; ink-2 batch calls still fail locally before network I/O. The 2026-10-18 pass: Only ink-whisper and dated ink-whisper-* IDs are batch-only. Other current and future IDs use the existing duplex transcription protocol.
Factory argument requirements are defined by the public Swift factory signatures. Use Public symbols when you need the exact initializer or factory declaration.
Return to the provider matrix.