examples

A collection of working examples showing how to use Deepgram SDKs with popular platforms, frameworks, and ecosystems.

→ Contributing · → Open PRs · → Suggest an example

Examples

#	Example	Language	Integration
010	Getting Started — Transcribe a URL with Node.js	Node.js	Deepgram SDK
020	Twilio Media Streams — Real-Time Call Transcription	Node.js	Twilio
021	Twilio Voice + Deepgram Voice Agent — AI Phone Agent	Node.js	Twilio
030	LiveKit Agents — Voice Assistant with Deepgram STT	Python	LiveKit
040	LangChain STT Tool — Transcribe Audio in AI Pipelines	Python	LangChain
050	Vercel AI SDK — Transcribe Audio and Generate Speech with Deepgram	Node.js	Vercel AI SDK
051	Next.js Streaming STT + TTS with Deepgram via the Vercel AI SDK	Node.js	Vercel AI SDK
052	Vercel AI SDK Agent with Deepgram Voice Tools	Node.js	Vercel AI SDK
060	Discord Bot — Transcribe Audio Attachments with Deepgram	Node.js	Discord
070	Vonage Voice API — Real-Time Call Transcription	Node.js	Vonage
080	Pipecat Voice Pipeline — Conversational Bot with Deepgram STT & TTS	Python	Pipecat
090	Expo Live Transcription	Node.js	Expo
100	FastAPI Audio Transcription API	Python	FastAPI
110	Cloudflare Worker — Edge Audio Transcription	Node.js	Cloudflare
120	Slack Bot — Auto-Transcribe Audio Messages with Deepgram	Node.js	Slack
130	Telegram Voice Transcription Bot	Python	Telegram
140	Audio to Subtitles CLI	Python	Deepgram SDK
150	Flutter Voice Transcription	Dart	Flutter
160	LlamaIndex Audio Document Loader — Transcribe Audio into RAG Pipelines	Python	LlamaIndex
170	Electron Live Transcription Overlay	Node.js	Electron
180	Zoom Cloud Recording Transcription with Deepgram	Node.js	Zoom
190	Daily.co Real-Time Transcription Overlay	Node.js	Daily.co
200	Vanilla JavaScript Browser Transcription (No Bundler)	Node.js	Deepgram SDK
210	OpenAI Agents SDK Voice Pipeline with Deepgram STT & TTS	Python	OpenAI Agents SDK
220	Django Channels Real-Time Transcription with Deepgram Live STT	Python	Django
230	n8n Community Nodes for Deepgram	Node.js	n8n
240	Nuxt Streaming STT + TTS with Deepgram	Node.js	Nuxt
260	NestJS WebSocket Real-Time Transcription	Node.js	NestJS
270	SvelteKit Real-Time Live Transcription	Node.js	SvelteKit
280	Express.js + React Live Transcription (TypeScript)	Node.js	Express + React
290	AWS Lambda Serverless Audio Transcription	Python	AWS Lambda
300	Spring Boot Real-Time Transcription with Deepgram	Java	Spring Boot
310	CrewAI Voice-Enabled Multi-Agent System with Deepgram	Python	CrewAI
340	Tauri Desktop Live Transcription	Python	Tauri
350	Asterisk / FreeSWITCH PBX to Deepgram Streaming STT	Python	Asterisk/FreeSWITCH
360	Kotlin Android Live Transcription	Kotlin	Jetpack Compose
370	Swift iOS Live Transcription	Swift	SwiftUI
400	Riverside.fm Recording Transcription	Node.js	Riverside
420	SignalWire Real-Time Call Transcription with Deepgram STT	Node.js	SignalWire
430	Telnyx TeXML Stream to Deepgram Real-Time Transcription	Node.js	Telnyx
440	Plivo Audio Streaming — Real-Time Call Transcription	Node.js	Plivo
450	Jitsi Meet Real-Time Transcription with Deepgram STT	Node.js	Jitsi
460	Webex Recording Transcription with Deepgram	Node.js	Webex
480	OBS Studio Live Captioning Plugin (C + Deepgram STT)	Python	OBS Studio
490	Haystack Audio Transcription Pipeline with Deepgram STT	Python	Haystack
500	Semantic Kernel Voice Plugin with Deepgram (.NET)	JavaScript	Semantic Kernel
510	Gin Real-Time WebSocket Transcription Server	Go	Gin
520	Deepgram Proxy Server (Node.js)	Node.js	Deepgram SDK

CI / testing

Every PR that touches examples/** runs language-specific test jobs automatically. The e2e-api-check status check is required before merge.

Language	Marker file
Node.js / TypeScript	package.json
Python	requirements.txt or pyproject.toml
Go	go.mod
Java	pom.xml or build.gradle
Rust	Cargo.toml
.NET	*.csproj or *.sln
Dart	pubspec.yaml
Kotlin	build.gradle.kts
Swift	Package.swift or *.xcodeproj
CLI	example.sh or src/*.sh

All examples are also tested on a recurring schedule to catch regressions from SDK updates or API changes.

Automated example building

Suggesting a new integration is as simple as opening an issue. From there, a GitHub Actions workflow handles the build: an LLM agent (Claude Opus) writes the code, runs it in a Docker sandbox, diagnoses failures, and iterates until tests pass — then opens a PR for human review before anything merges. The workflow also sweeps open suggestions on a schedule, so nothing sits unbuilt indefinitely.

The agent uses a neurosymbolic architecture to stay on track. A symbolic working memory records observed facts as tools fire — which files exist, whether tests are passing, which phases are complete — without the LLM being able to misreport them. A forward-chaining rule engine pattern-matches tool output each turn and injects targeted guidance when things go wrong: missing imports get named, repeated failing commands trigger a strategy nudge, turn-budget warnings reprioritise the remaining work. And a constraint checker holds a formal veto over completion: when the agent signals it's done, the checker verifies the definition of done deterministically — required files exist, test directories are non-empty, no hardcoded secrets — and rejects the signal if anything is missing. The LLM handles open-ended reasoning and generation; the symbolic layer handles state, rules, and verification.

Directory structure

examples/
{NNN}-{slug}/ # e.g. 010-getting-started-node
README.md # Quickstart — what it does, prerequisites, env vars, how to run
BLOG.md # Developer walkthrough — how it was built, step by step
.env.example # Every required environment variable (no values)
screenshot.png # 1240×760 Playwright screenshot (UI/terminal examples)
src/ # Source code
tests/ # Tests — exit 0=pass, 1=fail, 2=missing credentials
tests/
e2e.py # Deepgram STT + TTS smoke test (runs on every PR)
.github/
workflows/ # CI workflows
scripts/ # Agent scripts (plan, build, filter secrets, symbolic layer)
ISSUE_TEMPLATE/ # Issue templates

Numbering convention

Examples are numbered globally in increments of 10: 010, 020, 030 … A platform owns its group — a second Twilio example would be 021, not a new slot. New platforms claim the next free multiple of 10.

Setup

1. Add DEEPGRAM_API_KEY as a repository secret — required for E2E tests
2. Add partner credentials as needed (each example's .env.example lists them)

GITHUB_TOKEN is provided automatically by GitHub Actions.

Contributing

Open an issue to suggest a new example, or submit a PR directly. See CONTRIBUTING.md for guidelines.