Skip to main content
Cartesia provides the voice that speaks your agent’s responses, and transcription of what the caller says. One connection covers both. Cartesia is built for streaming, and on a phone call that shows up as shorter gaps before the agent starts speaking.

What you need

One key covers both services.

Getting your API key

  1. Sign in at play.cartesia.ai/keys.
  2. Create a new key.
  3. Copy it and store it somewhere safe.
Keys begin with sk_car_.
For anything running in production, create the key under an account that will outlast any individual rather than a personal login.

Adding the connection

  1. Go to Connections and choose Cartesia.
  2. Paste the API key.
  3. Choose which services this connection drives — voice, transcription, or both.
  4. Save.
Saving stores the key; it is not checked against Cartesia at that moment. A wrong or expired key first shows up on a live call, so place a test call after connecting.

What you can select afterwards

Choose your models on Model Configurations. Voice — pick from the dropdown, or enter a voice ID manually. Cartesia IDs are UUIDs like 3faa81ae-d3d8-4ab1-9e44-e50e46d33c30, not names. Browse the library at play.cartesia.ai/voices. Language — two-letter ISO 639-1 codes (en, fr, de). A voice keeps its original accent in other languages. Transcription coverage is narrower than voice — check the speech-to-text documentation. Cloned voices — a clone gets an ID in the same UUID form. It belongs to the account that created it, so use a key from that account.

Using Cartesia for only one service

The connection does not have to drive both roles. Turn off the one you do not want, and SentiVue Talk uses your other connections for it — Cartesia voices with transcription from Deepgram, or the reverse. Change this at any time on the connection.

Completing the pipeline

Cartesia has no language model. You also need an LLM provider to decide what the agent says, added as a separate connection.

Troubleshooting

Saving does not verify the key with Cartesia. Confirm you copied the whole key including the sk_car_ prefix, and that it has not been deleted or rotated in the Cartesia playground. Place a test call after any key change.
Cartesia plans include a monthly credit allowance, consumed by voice and transcription usage. Check the usage dashboard in the Cartesia playground and top up or move to a larger plan.
Cartesia voice IDs are UUIDs. Copy the ID from the voice library rather than the voice’s display name. For a cloned voice, confirm the key belongs to the account that owns the clone.
Check the language first. Cartesia’s transcription covers far fewer languages than its voices, and an unsupported language is transcribed as though it were a supported one rather than failing outright. If your callers’ language is not covered, keep Cartesia for voice and move transcription to another transcriber.
Cartesia pins each request to a dated API version. When it retires a version, integrations still on it stop working. There is nothing to set on the connection — check Cartesia’s changelog for deprecation notices and contact support if a working agent fails with no other explanation.