AI Agent Index
ByHeather MacAvelia·Independently reviewed·Published Nov 21, 2026·Updated Nov 21, 2026
Independently verified against live vendor data on Nov 21, 2026.

Developer API from Deepgram that runs real-time voice agents over one WebSocket, combining speech recognition, an LLM and speech synthesis. $200 free credit, then from $0.075/min (USD).

How we scored it

Autonomy

4/5

Integrations

4/5

Pricing clarity

4/5

Evidence

5/5

Setup

4/5

The facts

Deepgram Voice Agent API is Deepgram's developer API for real-time voice agents. Deepgram is best known for its speech-to-text and text-to-speech models, and the Voice Agent API combines them with a language model in one WebSocket connection: it listens, detects the end of each turn, handles interruptions, calls functions the developer defines and speaks the reply. Prompts, models and voices can be changed mid-conversation, and the API falls back to another LLM or voice automatically if one fails. Think Live, a Deepgram model that runs alongside the reasoning LLM, manages turn-taking and short acknowledgements. Deepgram says more than 1,000 organizations have built on the Voice Agent API and counts 47 million minutes of voice agent conversation.

The managed LLM step offers models from OpenAI, Anthropic, Google and NVIDIA. Teams can bring their own LLM through an OpenAI-compatible endpoint or Amazon Bedrock, and their own voice from OpenAI, ElevenLabs, Cartesia or Amazon Polly. Deepgram publishes Voice Agent guides for Twilio, Plivo, Genesys Cloud, Amazon Connect and AudioCodes, framework guides for LiveKit and Pipecat, SDKs for Python, JavaScript, Go, .NET, Java and Rust, and a browser agent SDK. Its speech models also connect to Zapier, Make.com, Zoom and Google Dialogflow CX. Endpoints run in Global, EU, India and Australia regions, and Enterprise customers can self-host on Kubernetes.

Prices are in US dollars and the Voice Agent API is billed per minute of WebSocket connection time. Pay as you go starts with a $200 free credit, needs no credit card and has no minimum, and the credit does not expire. On pay as you go the Standard agent costs $0.075 a minute, Standard with your own TTS $0.056, Custom with your own LLM $0.065, Custom with your own LLM and TTS $0.050, Advanced $0.163 and Advanced with your own TTS $0.122. Growth plans buy pre-paid annual credits from $4,000 a year and bring those rates to $0.068, $0.051, $0.059, $0.040, $0.146 and $0.110. Enterprise pricing is by quote.

The Standard rates cover the LLMs Deepgram places in its Standard tier, such as mini, nano and Haiku models. LLMs marked Advanced, such as gpt-5.4 and claude-sonnet-4-6, use the Advanced rates. On Growth, usage beyond the pre-paid credits is billed weekly at the Growth rate plus 10%, and concurrency is up to 60 voice agent connections in North America and 45 in Europe, Australia and India. Sessions close after 2 hours. Self-hosting and HIPAA BAAs need an Enterprise agreement, and Deepgram's own LLMs are in early access for contract customers only.

Deepgram Voice Agent API suits developer teams that want to build their own voice agent and control the LLM, voice and telephony behind it. Teams that want one platform deploying the same agent across phone, web chat, WhatsApp, SMS and email should look at ElevenAgents. Teams that want a cloud phone system with AI voice agents built in should look at Aircall AI, and service businesses that want a voice agent built from their website should look at Workforce Wave.

Current state Q4 2026: In October 2026 Deepgram announced that it is acquiring Daily, the company behind the open-source Pipecat framework, and introduced the Real-Time Context Standard, an open protocol. The India endpoint became generally available in September 2026. Deepgram's developer documentation lists SOC 2 Type 1 and Type 2, GDPR, CCPA, HIPAA with a BAA on request, PCI compliance and the Australian Privacy Principles, with FIPS 140-3 images for self-hosted deployments. By default Deepgram retains Voice Agent audio, transcripts and agent context to improve its models, and each request can opt out, which turns on zero data retention. G2 reviews cover Deepgram as a whole, since the Voice Agent API has no separate G2 profile.

Pricing

usage-based · $0.075 per minute

View pricing ↗

Segment

Business

Setup

moderate

Verified

Nov 21, 2026

Transparency

Mostly Public

Contract

Monthly or Annual

Data training

Opt-out

Source: developers.deepgram.com/trust-security/your-data, deepgram.com/terms · read Nov 21, 2026

Human in loop

Optional

Deepgram Voice Agent API pricing

Deepgram Voice Agent API pricing starts at $0.075 per minute. Monthly and annual billing are both offered. Checked against the vendor's own pages on Nov 21, 2026.

Deepgram Voice Agent API features

Voice agentAutonomousMultilingualTicket resolution

Deepgram Voice Agent API pros and limitations

Editorial assessment

Pros

  • ✓One WebSocket handles speech recognition, the LLM and speech synthesis, with barge-in, end-of-turn detection, function calling and automatic fallback.
  • ✓Teams can use managed LLMs from OpenAI, Anthropic, Google and NVIDIA or bring their own LLM and voice, on Global, EU, India or Australia endpoints or self-hosted.
  • ✓Every Voice Agent rate is published, and pay as you go starts with a $200 credit that needs no credit card and does not expire.

Limitations

  • –LLMs Deepgram marks Advanced, such as gpt-5.4 and claude-sonnet-4-6, raise the pay-as-you-go rate to $0.163 a minute, billed on connection time.
  • –Request audio, transcripts and agent context are retained to improve Deepgram's models unless each request sets the opt-out parameter.
  • –Building an agent takes developer work over the API or SDKs, and self-hosting and HIPAA BAAs need an Enterprise agreement.

Deepgram Voice Agent API technical details

Deployment
apicloudself-hosted
Model architectureDeepgram speech-to-text and text-to-speech models (Flux, Nova-3, Aura) with the customer's choice of LLM from OpenAI, Anthropic, Google or NVIDIA, or a bring-your-own LLM or TTS endpoint
Autonomous rateVoice agents run full conversations on their own: listening, detecting turns, calling developer-defined functions and speaking replies, with no human step unless the developer builds one. Deepgram does not publish a resolution or task-completion rate.
Integrations
ZapierZoomTwilioPlivoGenesys CloudAmazon ConnectAudioCodesLiveKitPipecatMakeDialogflow
Security
SOC 2 Type IIHIPAAGDPRCCPA

Alternatives to Deepgram Voice Agent API

Rating

4.3/ 5
★★★★★

Editorial score

How we score this →

Recognition

Listed 2026
Embed these on your site →

Built for

B2BEnterpriseDevToolsSaaS

Leave a review

Used Deepgram Voice Agent API? Tell other buyers how it worked for you. Takes about two minutes.

Write a review

Compare

Related Content

Is this your product?

Last checked Nov 21, 2026. AI assistants repeat what this page says, so keep it current: Featured Listings are re-checked every 14 days.

Get Featured →Claim it free

AI Agent Price & Rating Tracker

Free, every two weeks. Price changes, new agent launches, acquisitions, and rating updates across the AI agent market. Verified against live vendor data, not vendor marketing.

No spam. Unsubscribe anytime. We never share your email. Read past issues