Deepgram
Also known as: Deepgram AI
Voice AI infrastructure with real time speech to text (Nova, Flux), text to speech (Aura), and a Voice Agent API that unifies listening, thinking, and speaking.
Deepgram is a voice AI infrastructure platform that provides fast and accurate real time APIs for speech to text, text to speech, and voice agents, available in real time and batch, cloud and self hosted. It is best understood as a set of speech building blocks rather than one feature. Nova is its high performance speech recognition family, with Nova-3 leading on noisy call center audio and supporting more than forty five languages, and Flux is a conversational speech recognition model built for voice agents with turn taking and low latency. Aura is its text to speech family for natural, low latency voice output, and the Voice Agent API unifies speech to text, text to speech, and LLM orchestration into a single managed route for building voice to voice systems, with function calling and mid conversation prompt updates.
Deepgram is the specialist choice when voice needs to work inside a live application rather than after a recording finishes, and it is widely used in regulated and high stakes settings, with a Nova-3 Medical route tuned for medical terminology, plus add ons for speaker diarization, redaction, and keyterm prompting. Deployment spans cloud, self hosted, on premise, VPC, and single tenant, with an EU endpoint, and the platform carries SOC 2 Type II, HIPAA with a business associate agreement for enterprises, GDPR, CCPA, and PCI compliance.
Pricing is usage based and detailed. A two hundred dollar free credit that never expires opens all endpoints with no card, then Pay As You Go bills per minute for speech to text, per thousand characters for Aura, and per minute of WebSocket time for the Voice Agent API, with bring your own LLM and TTS options lowering the agent rate. A Growth plan starting around four thousand dollars a year gives discounted prepaid rates, and Enterprise is custom for self hosted deployment, compliance, and support.
Vendor details
Canonical URL
https://deepgram.com
Category
Agent infrastructure
Subcategory
Voice infrastructure
Funding status
Independent voice AI platform, founded 2015. Widely adopted for low latency speech infrastructure, with SOC 2 Type II, HIPAA, GDPR, CCPA, and PCI compliance and self hosted deployment for enterprises.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Real time and batch APIs with SDKs and WebSocket streaming for speech to text, text to speech, and voice agents. The Voice Agent API unifies speech to text, LLM orchestration, and text to speech in one managed route with function calling and bring your own LLM or TTS. Add ons cover diarization, redaction, and keyterm prompting, with an EU endpoint for residency.
In practice
Your voice agent needs accurate transcription on noisy call center audio. Deepgram Nova-3 leads on challenging audio with sub 300 millisecond streaming, so the agent hears callers correctly in real time.
You want to build a voice to voice agent without stitching components. The Voice Agent API unifies speech to text, LLM orchestration, and text to speech in one route with function calling and bring your own LLM.
A healthcare deployment needs compliant, on premise voice. Deepgram self hosts with HIPAA and a business associate agreement, a Nova-3 Medical route for terminology, and redaction for sensitive data.
Sources & related URLs
Capability coverage
6.0 / 14 capabilities · 43%
| Integrations & Tool CallingAPIs and SDKs, Voice Agent API with function calling, telephony and LLM integration, docs 2026-07-06 | Full |
|---|---|
| Workflow OrchestrationVoice Agent API unifies speech to text, LLM orchestration, and text to speech in one route, docs 2026-07-06 | Partial |
| Knowledge Grounding & RAGVoice layer, no knowledge grounding or RAG | Unable to verify |
| Human Oversight & GuardrailsRedaction removes PII but there is no agent guardrails or human oversight product | Unable to verify |
| Security, Identity & GovernanceSOC 2 Type II, HIPAA with BAA, GDPR, CCPA, PCI, self hosted and EU endpoint, redaction, docs 2026-07-06 | Full |
| Observability & AuditabilityAudio Intelligence analytics (summarization, sentiment, topics) and usage tracking, docs 2026-07-06 | Partial |
| Memory & State PersistenceMid conversation context but no persistent memory layer | Unable to verify |
| Deployment & Data ResidencyCloud, self hosted, on premise, VPC, single tenant, EU endpoint for residency, docs 2026-07-06 | Full |
| Prebuilt Agents, Templates & PacksPrebuilt and industry tuned models exist but no prebuilt agents | Unable to verify |
| Triggers & Channel CoverageVoice and telephony channel via the Voice Agent API and WebSocket sessions, docs 2026-07-06 | Partial |
| Model Flexibility & RoutingOwn STT and TTS model families plus bring your own LLM and TTS in the Voice Agent API, no cross provider routing, docs 2026-07-06 | Partial |
| APIs, SDKs & MCP ExtensibilityAPIs, SDKs, WebSocket streaming, Voice Agent API with function calling, docs 2026-07-06 | Full |
| Testing, Debugging & OptimizationCustom model training and production audio testing but no eval product | Unable to verify |
| Browser & Computer UseNo browser or computer use | Unable to verify |
Pricing
Free $200 credit; Pay as you go; Growth from about $4,000/year; Enterprise
usage (minutes, characters, agent minutes)
Included quota
A $200 free credit that never expires opens all endpoints (STT, TTS, Voice Agent API, Audio Intelligence) with no card. Pay As You Go has no minimums. Growth (from about $4,000 a year) gives discounted prepaid credits and higher concurrency. Enterprise adds negotiated rates, self hosted deployment, and dedicated support.
What is public
Detailed per minute and per character rates, add on rates, free credit, and Growth minimum are published; Enterprise is custom.
Billing mechanics
Separate usage meters per product: speech to text per minute, Aura text to speech per 1,000 characters, and the Voice Agent API per minute of WebSocket connection time, with bring your own LLM or TTS lowering the agent rate. Add ons like diarization and redaction meter separately.
Cost watchouts
Add ons (diarization, redaction, summarization) are billed separately and can raise effective cost meaningfully. Multichannel audio multiplies base cost by channel count. The Growth plan carries a roughly $4,000 annual minimum, and in late 2025 Aura-2 pricing doubled and STT rates rose.
Variable cost rationale
Cost is entirely usage across minutes, characters, and agent session time, amplified by separately metered add ons and multichannel multipliers, so spend tracks call and audio volume directly.
Overage / add-ons
Pay As You Go bills per minute (STT), per 1,000 characters (Aura), and per minute of WebSocket time (Voice Agent API), with add ons metered separately.
Sales call required
Mixed (some tiers require a call)
Free / trial
$200 free credit, never expires, no credit card, across all endpoints
Lowest paid plan
Pay As You Go usage after the free credit; Growth from about $4,000 a year
Commercial notes
Independent, founded 2015. SOC 2 Type II, HIPAA with BAA, GDPR, CCPA, and PCI compliant, with self hosted and EU endpoint options. Nova-3 Medical route for healthcare.
Key ambiguities
Promotional streaming rates appear on the public page, so verify current live pricing; effective cost depends heavily on model choice and enabled add ons.
Related vendors
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- Apify — Cloud platform for web scraping and automation with 45,000+ prebuilt…
- Arcade — Authenticated tool calling platform and MCP runtime that handles…
- Arize AI — AI observability and evaluation platform that traces, evaluates, and…
- Beam — Open source serverless sandboxes and GPU cloud for production…