Providers

Wireface Chat doesn't include an AI model: you connect your own accounts. Add keys in the admin panel under Providers (or set ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY or ELEVENLABS_API_KEY before the first-run setup).

Provider	Text replies	Realtime voice	Speech to text	Text to speech	Notes
Anthropic	yes (Claude)				Voice through the cascade
OpenAI	yes	yes (Realtime API)	yes	yes
Google Gemini	yes	yes (Live API)		yes	Live takes a camera video stream
ElevenLabs		ElevenLabs Agents	yes (Scribe)	yes	Thousands of voices

How keys are handled

• Checked first. The server tries a key before saving it, and refuses one the provider rejects.
• Encrypted at rest with AES-256-GCM under the server's master key (DATA_DIR/master.key or WIREFACE_MASTER_KEY). See Backups.
• Never sent to a browser. The admin panel only ever shows the last four characters. All provider calls, including voice, go from your server.
• Status. Each connection shows ok, invalid (rejected), quota (out of credit), error (couldn't check) or unchecked. If a key starts failing during chats, the server marks it, and the bot's editor shows the problem. Test checks it again; you can replace the key without touching the bots that use it.

Models and voices

Once a key is connected, the server asks the provider which models (and voices) it can use and fills the admin panel's lists. It sorts them by what they're for (text, realtime, speech to text, text to speech) and marks a recommended one of each, which new bots start with. The lists are cached for a day; refresh them from the connection.

Which to choose

• Text only: any of Anthropic, OpenAI or Gemini.
• Fast, natural voice: OpenAI Realtime or Gemini Live (speech to speech).
• Claude, with a voice: Anthropic for the brain, plus the cascade: speech to text from OpenAI or ElevenLabs, and a voice from ElevenLabs, OpenAI or Gemini.
• A particular voice (a cloned or designed one): ElevenLabs, through the cascade or ElevenLabs Agents.
• A live camera view: Gemini Live. The other engines see camera frames per message or on request.

See Voice for how the engines differ.

Costs

Usage is billed by each provider to your account. The admin panel estimates spend per bot from built-in list prices (unknown models count as free), and the bot's security.caps can stop it at a daily limit. See Configuration.

Corporate gateways and proxies: point the server at them with the PROVIDER_BASE_URL_* and PROVIDER_WS_URL_*environment variables.