recipes

Agent-maintained recipes showing how to use every Deepgram SDK feature across every supported language.

All recipes are built and kept current by autonomous agents. Humans can direct, override, and add recipes at any time.

→ Contributing · → Open PRs · → Coverage matrix · → Request a recipe

How it works

1. Discover — Hourly: agents scan SDK repos for new releases and compare against recipes/ to find coverage gaps, creating queue issues for missing recipes
2. Generate — Agents pick up queue issues, fetch current SDK docs via Kapa, and write runnable example.* + example_test.* + README.md for each missing recipe, raising a PR
3. Test — Language-specific CI workflows run every example against the real Deepgram API
4. Merge — PRs merge once tests pass
5. Update — Coverage matrix rebuilds automatically on merge
6. Review — Weekly: agents check SDK release notes for new features and API changes, updating features.json and queuing fixes for stale recipes

Recipe structure

recipes/{language}/{product}/{version}/{recipe}/
  example.{ext}       # runnable, < 50 lines, reads DEEPGRAM_API_KEY from env
  example_test.{ext}  # runs the example as a subprocess, asserts output
  README.md           # what the feature does · params · sample output · how to run

Languages

Language	SDK	Install
Python	deepgram-python-sdk	pip install deepgram-sdk
JavaScript	deepgram-js-sdk	npm install @deepgram/sdk
Go	deepgram-go-sdk	go get github.com/deepgram/deepgram-go-sdk/v3
.NET	deepgram-dotnet-sdk	dotnet add package Deepgram
Java	deepgram-java-sdk	Maven: com.deepgram.sdk
Rust	deepgram-rust-sdk	deepgram = "0.6"
CLI	deepgram/cli	pip install deepctl

Products

Product	API versions	Example recipes
Speech-to-Text	v1 (nova-3, nova-2, nova, enhanced, base) · v2 (flux-general-en)	transcribe-url, transcribe-file, paragraphs, diarize, smart-format, utterances, summarize, sentiment, topics, intents, detect-entities, detect-language, redact, search, keywords, streaming, …
Text-to-Speech	v1 (aura-2 voices)	generate-audio, stream-audio, websocket-streaming, select-model, select-encoding
Audio Intelligence	v1	summarize, sentiment, topics, intents, entities
Voice Agents	v1	connect, custom-llm, custom-tts, function-calling

Recipes

Last updated 2026-03-29 20:49 UTC

	Python	JavaScript	Go	.NET	Java	Rust	CLI
Speech-to-Text `v1`	11/19	✅	✅	✅	✅	❌	❌
Speech-to-Text `v2`	❌	✅	✅	✅	✅	❌	❌
Text-to-Speech `v1`	1/5	✅	✅	✅	✅	❌	❌
Audio Intelligence `v1`	❌	✅	✅	✅	✅	❌	❌
Voice Agents `v1`	❌	✅	✅	✅	✅	❌	❌
Total	12/35	35/35	35/35	35/35	35/35	0/35	0/35
Tests	—	—	—	—	—	—	✅

Agent schedules

Workflow	Schedule	Purpose
pm	Every hour at :07	Find coverage gaps → create queue:generate issues
engineer	Every hour at :27	Pick up queue issues → generate recipes → open PRs
researcher	Saturdays 08:17 UTC	Check SDK release notes for new features and API changes
lead-coverage	On PR merge + every 6 h	Rebuild COVERAGE.md and per-language READMEs
lead-reconcile	Daily 11:45 UTC	Verify README counts; flag incomplete recipe directories

Setup

1. Add ANTHROPIC_API_KEY as a repository secret
2. Add DEEPGRAM_API_KEY as a repository secret
3. Add KAPA_API_KEY as a repository secret (used by agents to query live Deepgram docs)
4. Set KAPA_PROJECT_ID as a repository variable: 1908afc6-c134-4c6f-a684-ed7d8ce91759
5. Run Actions → Setup Labels → Run workflow once to create all issue labels
6. Enable auto-merge: Settings → General → Pull Requests → Allow auto-merge
7. Set branch protection on main: require status checks → add E2E / test as the required check — this runs on every PR, makes real API calls, and is the gate for auto-merge. The per-language test-* workflows also run when their files change but are not required (they fail-fast and block human review if broken)

GITHUB_TOKEN is provided automatically by GitHub Actions.

Running agents locally

# Requires: ANTHROPIC_API_KEY set, DEEPGRAM_API_KEY set, gh auth login, git configured
claude --model claude-opus-4-6 -p "$(cat instructions/pm.md)"
claude --model claude-opus-4-6 -p "$(cat instructions/engineer.md)"

See instructions/README.md for all available instructions.