Cut your agent's biggest cost.
Speech, transcription, and chat on one key. Move your voice stack to mserve and keep every other line of code.
pip install elevenlabs−from elevenlabs import ElevenLabsclient = ElevenLabs(api_key=os.environ["MSERVE_API_KEY"],
)stream = client.text_to_speech.convert_as_stream(voice_id=voice_id, text=reply,)
Works where you build.
Any framework with a base URL setting, including Pipecat and LiveKit Agents. OpenAI Realtime over WebSocket included.
Streams as it speaks.
Audio for the first sentence starts while the rest renders, so turns stay natural.
Your bill, after.
Voice cloning, per million characters.
Voice cloning: $82.50 becomes $6.00. You save 93%.
Per 1M charactersElevenLabsmserve · Chatterbox
Your model on the same key.
Run the agent's reasoning on Qwen3.8 27B and its ears on Whisper through the same account.
userSummarize this call in one line.
assistantCustomer asked to move delivery to Thursday; confirmed and sent the updated receipt.
Move the biggest line item first.
One email when your key is ready. Nothing else.