Cartesia

Create responsive, natural-sounding speech in 44 languages, with voice customization and flexible deployment for applications and production voice agents.

Help voice agents understand speech accurately and respond quickly, with built-in turn detection and deployment options for production environments.

Automate inbound and outbound calls with voice agents that complete tasks, connect to business systems, and can be tested before launch.

Use AI assistants to generate speech, transcribe recordings, and manage voices and pronunciations through chat.

Sonic

Cartesia Sonic is a real-time text-to-speech model that generates expressive, natural-sounding speech for applications and production voice agents. It streams audio with model latency under 90 ms and supports 44 languages from one model. Developers can clone and localize voices, customize pronunciations, and access speech generation through an API. Sonic-3.6 can also be deployed on-premises, in a customer VPC, or through OEM licensing under enterprise contracts.

• Start Playback Before Generation Finishes
  Begin playing generated speech while the rest of the response is still being produced, supporting responsive interactions such as live voice conversations.
• Deliver More Natural Conversations
  Use context-aware delivery, natural pacing, and emotional expression to create speech that sounds more like a human conversation.
• Reach Listeners In 44 Languages
  Generate speech across languages and regional accents from one model to serve international audiences with native-quality voices.
• Create Consistent Custom Voices
  Create a voice clone from a short audio sample and use it to maintain a consistent brand voice. Cloning requires permission and verified speaker consent.
• Deploy In Your Preferred Environment
  Choose API access, on-premises deployment, a customer VPC, or OEM licensing to support different production and infrastructure requirements.
• Preserve Voice Identity Across Languages
  Localize audio for other languages while retaining the original speaker's identity, tone, and emotion.
• Control Pronunciation Of Key Terms
  Set pronunciations for names, specialized terminology, and other words that need to sound exactly as intended.
• Speak Numbers And Terms Accurately
  Render alphanumeric sequences and heteronyms more accurately, helping speech generation handle details such as phone numbers, addresses, and confirmation codes.
• Meet Enterprise Security Requirements
  Use a product presented with enterprise compliance credentials and deployment options that can support data residency, sovereignty, and network isolation requirements.
• Reduce Costs And Support More Requests
  Use the computational efficiency of state space models on long sequences to support lower inference costs and higher concurrency.

Ink

Ink is Cartesia’s streaming speech-to-text model, designed for real-time conversations and production voice agents. Ink-2 delivers low word error rates, transcribes structured information such as phone numbers, dates, emails, currencies, and UUIDs, and is built for telephony, background noise, and varied accents. Native turn detection identifies when speakers start and finish, with semantic endpointing and eager turn-ending options. Ink-2 can be deployed on-premises, in a customer VPC, or through OEM licensing, and supports English, Spanish, French, Hindi, and Japanese.

• Transcribe Speech Accurately
  Improve the quality of voice-agent interactions with accurate streaming transcripts, including structured details such as phone numbers, dates, emails, currencies, and UUIDs.
• Detect Speaker Turns Natively
  Identify when a speaker starts and finishes without integrating or maintaining a separate turn-detection model, while reducing the risk of responding during a pause in mid-thought.
• Deliver Transcripts In 0.1 Seconds
  Support responsive voice-agent conversations with fast final transcripts and an eager turn-ending option that can reduce the interval between the speaker’s last word and the agent’s first response.
• Deploy On-Premises, In A VPC, Or Via OEM
  Choose on-premises, customer VPC, or OEM deployment options for production requirements, including data sovereignty and residency needs.
• Transcribe Speech In Noisy Settings
  Transcribe audio in real-world settings with background noise without adding a separate audio-filtering model.
• Increase Throughput And Reduce Compute Costs
  Support higher throughput with lower compute costs at scale, while maintaining quality as usage grows.
• Meet Enterprise Compliance Requirements
  Use a product listed as HIPAA compliant and aligned with SOC 2 Type 2, GDPR, and PCI requirements, with air-gapped on-premises deployment available.
• Transcribe Speech In Five Languages
  Build transcription workflows for English, Spanish, French, Hindi, and Japanese.

Managed Agents

A managed platform for building and deploying AI voice agents for inbound and outbound phone calls. Cartesia handles listening, turn-taking, interruptions, tool calls, and knowledge retrieval, while a customer-selected AI model determines each reply. Agents can connect to business systems, use Cartesia or existing phone numbers, and be tested and evaluated before handling real calls.

Category: Conversational AI Platforms

• Coordinate The Full Voice Interaction
  Run key parts of a voice interaction in one managed system, from listening and conversation flow to tool calls and knowledge retrieval.
• Handle Calls In Both Directions
  Use agents to answer incoming calls or place outbound calls, including batch campaigns to lists of up to 5,000 numbers.
• Connect Agents To Business Systems
  Connect agents to configured tools and business systems so they can use relevant information and perform tasks during calls.
• Test And Score Calls Before Launch
  Use browser-based test calls and custom metrics to assess agent performance before customers interact with the agent.
• Meet Enterprise Security Requirements
  Support enterprise compliance requirements with listed HIPAA, SOC 2 Type 2, and GDPR credentials, and use on-premise deployment options described by Cartesia.
• Support Natural Phone Conversations
  Let callers speak naturally, including pausing or interrupting, while the agent manages conversation turns.
• Connect Existing Phone Numbers
  Route calls through Cartesia-provisioned numbers or connect existing numbers and supported phone providers to agents.
• Review Every Recorded Interaction
  Inspect recordings, transcripts, summaries, and evaluation results to understand how calls were handled.
• Choose The Agent's AI Model
  Select the AI model that decides an agent's replies and change it as needed.
• Launch Agents From Templates
  Start with a template, configure its instructions and connections, then test and publish the agent for use on a phone number or web page.
• Serve Callers In 44 Languages
  Let callers communicate with voice agents in any of the 44 languages Cartesia lists for its agents.
• Pay For Agent Minutes Used
  Use per-minute pricing for agent calls, with separate charges listed for Cartesia phone-number telephony and AI model usage.

Cartesia MCP server

Cartesia’s hosted Model Context Protocol server gives supported AI assistants access to tools for generating speech, transcribing audio, and working with voices and pronunciation dictionaries. Connect from Claude Code, Cursor, ChatGPT, or Grok, then use these tools in chat instead of writing API calls. Sign in with a Cartesia account to use the hosted server; tool calls use Cartesia credits.

• Connect Your AI Assistant
  Use Cartesia tools from supported AI assistants, with setup instructions for Claude Code, Cursor, ChatGPT, and Grok.
• Generate Speech From Text
  Generate speech from assistant prompts with configurable delivery and pronunciation, then retrieve the saved audio through a download link valid for 24 hours.
• Transcribe Audio Recordings
  Submit audio for batch or streaming transcription and use the resulting text in assistant workflows, such as searching and summarizing recordings.
• Sign In With Your Cartesia Account
  Connect to the hosted server by signing in with a Cartesia account; the server creates an organization MCP API key if needed.
• Find And Inspect Voices
  Search voices using available filters, including language and gender, and retrieve details for a selected voice.
• Clone And Manage Voices
  Create a voice clone from a recording, maintain its name or description, and remove cloned voices when needed.
• Create And Manage Pronunciations
  Build and maintain pronunciation dictionaries for names, product terms, or other words that need specific pronunciations in generated speech.
• Adapt Voices To Other Languages
  Localize a voice for another language or dialect while working through an AI assistant.
• Explore Models And Manage Accents
  Inspect available accents and speech models, then add catalog accents to or remove them from an instant voice clone.
• Retrieve Cartesia Files
  Fetch a Cartesia file using its ID and receive a link to download it.
• Review Credit Usage
  Let administrators check Cartesia credit usage over time through an assistant.