Deepgram

Documented custom tool support. Voice Agent function calling lets the customer define functions in Settings; each runs client side in the customer's application or server side, where Deepgram calls a web endpoint the customer provides, so an agent can book appointments, send emails or update records in the customer's CRM, ERP or internal APIs during a live call.

SourceDeepgram, developers.deepgram.com (Function Calling)read 2026-09-27

A fixed pipeline that Deepgram runs end to end. Each Voice Agent session chains listen, think and speak, with an ordered LLM fallback chain per request and mid call Update messages for the prompt and providers; the buyer configures the pieces, not the control flow. The multi agent architecture guide sequences qualifier, advisor and closer agents through a CallOrchestrator in a reference repository the customer runs, so that orchestration is the customer's code.

SourceDeepgram, developers.deepgram.com (LLM Models, Build a Multi-Agent Architecture)read 2026-09-27

No retrieval structure over the customer's content. The documentation index lists speech to text, text to speech, audio and text intelligence, and Voice Agent pages, and none describes ingesting, indexing or retrieving the customer's documents; context reaches the agent through the prompt, loaded history and function results the customer supplies. Keyterm prompting tunes transcription vocabulary, not knowledge.

SourceDeepgram, developers.deepgram.com (llms.txt, Maintaining Context)read 2026-09-27

A customer controlled constraint on agent actions, with no approval step. Per function, defer_until_eot holds a call whose side effect cannot be undone (ending a call, spending money, sending a message) until speech to text confirms the caller has finished the turn, while read only functions still dispatch early; Deepgram's docs tell builders to set it on any irreversible action. That is a configurable constraint on what the agent may do; no person approves an action inside Deepgram.

SourceDeepgram, developers.deepgram.com (Function Calling: Irreversible Actions and Turn Confirmation)read 2026-09-27

Access and compliance are both documented. On access, there is a named access model of account permissions plus owner, admin and member project roles, each implying a listed set of scopes, with API keys created per project and scoped to those permissions. On compliance, Deepgram lists SOC 2 Type 1 and Type 2 by an independent auditor, certificates on request, HIPAA business associate with a BAA for Enterprise, GDPR, CCPA and PCI with a yearly review.

SourceDeepgram, developers.deepgram.com (Working With Roles & API Scopes, Data Privacy Compliance) and deepgram.com/pricingread 2026-09-27

Request logs and usage, with the per session trace left to the customer. The Console shows usage and up to 90 days of request logs, and a Usage API exports them to tools such as Grafana or Datadog. For the Voice Agent, Deepgram's own guide says the dashboard does not give per session, turn by turn observability and there is no separate logging API: the customer taps the WebSocket and persists the transcript, function call, latency and error events it emits. Events a vendor emits for the customer to store are not a trace the buyer can inspect on the vendor's side.

SourceDeepgram, developers.deepgram.com (Session Observability, Logs & Usage Data)read 2026-09-27

Conversation state lives inside a session. The agent keeps the prompt, turns, injected messages and function results as working memory for the call, and History (on by default) lets a new session load prior turns and function calls through agent.context.messages. The prior turns are stored and supplied by the customer's application; Deepgram documents no memory layer of its own with a stated scope and lifetime.

SourceDeepgram, developers.deepgram.com (Maintaining Context, History)read 2026-09-27

Named regions offered as an option and a customer environment option. Regional endpoints at api.eu.deepgram.com (EU, never routed outside it) and api.au.deepgram.com (Australian infrastructure for storage and inference) serve speech to text, text to speech and the Voice Agent with the same keys; self hosting runs in customer requisitioned cloud instances such as AWS or GCP, or the customer's data center, for Premium customers. A third party LLM in the think step runs on that provider's infrastructure.

SourceDeepgram, developers.deepgram.com (Regional Endpoints, Deployment Options, Data Privacy Compliance)read 2026-09-27

No packaged agents a buyer adopts. The Voice Agent template apps are one starter ported to twelve languages and frameworks on GitHub, which is sample code, evidence of a framework rather than of a ready solution; Reusable Agent Configurations store the customer's own agent blocks; industry tuned models such as Nova-3 Medical and Pharma are models, not agents.

SourceDeepgram, developers.deepgram.com (Template Apps, Reusable Agent Configurations)read 2026-09-27

Voice channel coverage without an autonomous wake. The Voice Agent works on phone calls through documented inbound and outbound telephony builds (Twilio, Amazon Connect) and in web pages through the Browser Agent SDK and an embeddable widget. Every session starts when a caller dials in or the customer's own server opens the socket or places the call (the outbound reference takes a POST from the customer's CRM or CLI), so Deepgram owns no event, schedule or webhook wake.

SourceDeepgram, developers.deepgram.com (Build an Outbound Telephony Agent, Browser Agent SDK)read 2026-09-27

The customer chooses the model. Each Voice Agent's Settings name the think provider and model from OpenAI, Anthropic, Google and NVIDIA models Deepgram manages, or the customer's own endpoint for Groq, Amazon Bedrock or any OpenAI compatible API, and an ordered array of providers acts as a per request fallback chain that can mix providers; listen and speak models are chosen the same way, including BYO TTS.

SourceDeepgram, developers.deepgram.com (LLM Models)read 2026-09-27

A documented API and SDKs for Deepgram's own platform. REST and WebSocket APIs for speech to text, text to speech, text intelligence and the Voice Agent, a management API for projects, keys and usage, an Agent Configuration API, Python, JavaScript, Go and.NET SDKs with a feature matrix, a Browser Agent SDK, and the dg CLI, whose built in MCP server proxies the developer API's tools.

SourceDeepgram, developers.deepgram.com (llms.txt, Reusable Agent Configurations, MCP Server)read 2026-09-27

No documented way to evaluate an agent change. The documentation index carries no evaluation harness, scored test cases or quality gate for voice agents; the testing pages validate integrations (a streaming starter kit, a SageMaker endpoint check). Reusable Agent Configurations name A/B testing voices or prompts as a use case, but the conversion or CSAT measurement is the customer's own. Custom model training improves transcription, which is model work, not agent evaluation.

SourceDeepgram, developers.deepgram.com (llms.txt, Reusable Agent Configurations)read 2026-09-27

The Browser Agent SDK embeds a voice agent in a web page, which is the product running in a browser, not an agent operating one, and telephony is a voice channel. No browser, desktop or computer control is documented.

SourceDeepgram, developers.deepgram.com (llms.txt, Browser Agent SDK)read 2026-09-27