ElevenLabs Voice Manager

The ElevenLabs page at /admin/elevenlabs provides voice library browsing, subscription usage monitoring, and a text-to-speech playground. It connects to the ElevenLabs REST API via xi-api-key header authentication.

Architecture

Admin UI → /api/admin/elevenlabs-manage → ElevenLabs REST API
                                           ├── /v1/voices (library)
                                           ├── /v1/models (model list)
                                           ├── /v1/user/subscription (usage)
                                           └── /v1/text-to-speech/:voiceId (generate)

All requests go through /api/admin/elevenlabs-manage. The generate-speech action returns raw audio/mpeg binary directly to the browser for playback.

Page Layout

Stats Row

Four stat cards:

Tier — subscription tier name (e.g., “free”)
Characters — used / limit with visual progress
Voices — number of available voices
Models — number of TTS models

Voice Library Tab

Grid of voice cards showing:

Voice name and category (e.g., “premade”, “cloned”)
Labels: accent, gender, age, description, use case
Preview button — plays the voice’s sample audio from ElevenLabs CDN

Voices are fetched from /v1/voices and include both premade library voices and any custom/cloned voices on the account.

TTS Playground Tab

Interactive speech generation form:

Text — textarea for the text to synthesize
Voice — dropdown populated from the voice library
Model — dropdown of available TTS models (default: eleven_turbo_v2_5)
Stability — slider (0.0 to 1.0, default 0.5) — lower = more expressive, higher = more consistent
Similarity Boost — slider (0.0 to 1.0, default 0.75) — higher = closer to original voice
Generate button with loading state
Audio player — appears after generation, plays the synthesized audio

The generated audio is returned as audio/mpeg binary and loaded into a <audio> element via URL.createObjectURL().

API Route

Route: /api/admin/elevenlabs-manage

Auth: Admin-only (CF Access JWT validation)

GET

Health check — returns connection status, subscription tier, character count, and character limit.

POST Actions

All POST requests take { "action": "...", ...params }:

Action	Parameters	Returns
`list-voices`	none	`voices[]` (id, name, category, labels, preview_url)
`list-models`	none	`models[]` (id, name, description, can_do_text_to_speech)
`get-subscription`	none	`tier`, `characterCount`, `characterLimit`, `voiceCount`, `voiceLimit`, `nextReset`
`generate-speech`	`text`, `voiceId`, `modelId` (optional), `stability` (optional), `similarityBoost` (optional)	Raw `audio/mpeg` binary

Error Handling

Missing API key → 500 with configuration error
Missing required params (text, voiceId) → 400
ElevenLabs API errors → 502 with error text
Health check and list actions have 5-10 second timeouts; speech generation has a 30-second timeout

Environment Variables

Variable	Required	Description
`ELEVENLABS_API_KEY`	Yes	ElevenLabs API key (`sk_...`)

Set in production via npx wrangler pages secret put. For local dev, add to .env or .dev.vars.

Free Tier Limits

The free tier provides 10,000 characters per month with monthly reset. Character count includes all text sent to the TTS endpoint — both from the admin playground and from agent tools (OpenClaw voice-call). Monitor usage via the stats row or the get-subscription action.

Voice Usage Across Arcturus-Prime

ElevenLabs voices are used in multiple places:

Feature	Location	How
TTS Playground	`/admin/elevenlabs`	Direct API call, audio returned to browser
Argonaut voice mode	`/admin/argonaut` chat	Browser STT (Whisper) → AI → ElevenLabs TTS → speaker
OpenClaw voice-call	Agent tool	Generates audio but currently falls back to Polly (needs public audio URL hosting)

The default voice is Adam (pNInz6obpgDQGcFmaJgB) with the eleven_turbo_v2_5 model.

Twilio Dashboard (/admin/twilio) — phone calls using Polly voices via TwiML
Argonaut Hub (/admin/argonaut) — voice chat mode using ElevenLabs TTS
OpenClaw Gateway (/admin/openclaw) — agent with voice-call tool capability