[kraenhansen]

Developers with an existing OpenAI Realtime client need a runnable example of pointing it at ElevenLabs Realtime, to go with the OpenAI Realtime API compatibility guide.

Merging this PR will:

• Add agents/nextjs/openai-realtime (PROMPT.md, setup.sh, README.md, generated example/): a Next.js voice demo using @openai/agents-realtime over WebSocket against wss://api.elevenlabs.io/v1/convai/realtime.
• Mint the short-lived client secret server-side in app/api/token with client.conversationalAi.realtime.createClientSecret.
• List the example in the root README.md.

---

Additional context intended for coding agents and bots

Origin

Ported from the Vite demo in elevenlabs/elevenlabs-dx#2577 onto templates/nextjs, following the repo's prompt-driven layout (modelled on agents/nextjs/guardrails).

Changes relative to the source demo

• Token minting uses the published SDK method (@elevenlabs/elevenlabs-js ≥ 2.74.0) instead of a raw fetch in a Vite dev middleware, and returns only { value } to the browser. The /api/token path matches the guide's browser snippet.
• Audio lifecycle is hardened: both wavtools AudioContexts are closed on teardown, the session tears down on transport connection_change: "disconnected" (not on every non-fatal error event), and player.connect() runs before the token fetch so Safari creates the playback context within the click's user activation.
• Removed internal-only knobs: local-backend URL env vars and feature-flag-specific error mapping.
• UI restyled to DESIGN.md (neutral, flat, template header preserved) with plain buttons instead of a shadcn Button and custom theme.

Known issue (not introduced here)

With pnpm 12, setup.sh exits with ERR_PNPM_IGNORED_BUILDS for template deps (sharp, msw, unrs-resolver) after installing packages; this applies to every Next.js setup.sh in the repo.

Test plan and verification

• tsc --noEmit (clean)
• eslint (clean)
• next build (succeeds; / static, /api/token dynamic)
• next start smoke test: / server-renders; /api/token without a key returns 500 {"error": "Server misconfiguration: ELEVENLABS_API_KEY is not set."}
• npx prettier . --check (matches CI)
• End-to-end voice session with a real ELEVENLABS_API_KEY, including Stop → Start cycles (verified manually)

🤖 Generated with Claude Code

[cursor]

PR Summary

Low Risk
Documentation and a self-contained example app; token route follows the same server-side secret pattern as other demos with no production service changes.

Overview
Adds a new OpenAI Realtime SDK with ElevenLabs Next.js example under agents/nextjs/openai-realtime, and links it from the root README.md.

The demo uses @openai/agents-realtime with OpenAIRealtimeWebSocket aimed at ElevenLabs’ OpenAI-compatible WebSocket (wss://api.elevenlabs.io/v1/convai/realtime). A GET /api/token route mints a short-lived client secret via createClientSecret and shared GA sessionConfig (PCM 24kHz, semantic_vad, audio output only) so ELEVENLABS_API_KEY stays server-side.

The client hook streams mic PCM with wavtools (SDK audio is WebRTC-only), plays agent audio, builds an ordered transcript from transport events, and tears down on disconnect with explicit AudioContext cleanup and Safari-friendly player.connect() ordering. PROMPT.md and setup.sh scaffold example/ from templates/nextjs like other agent samples.

^{Reviewed by Cursor Bugbot for commit e6006e0. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 7649ff0. Configure here.}