.env.example

# Text: gemini (default) or openai. Used for chat, reports and voice reasoning.
MING_TEXT_PROVIDER=gemini
GEMINI_API_KEY=
# Optional model overrides:
# GEMINI_MODEL=gemini-3.8-flash
# OPENAI_TEXT_MODEL=gpt-6-luna

# Voice: openai (default OpenAI Live) or deepgram (hold-to-talk STT/TTS).
MING_VOICE_PROVIDER=openai
# Needed for OpenAI Live, or if MING_TEXT_PROVIDER=openai.
OPENAI_API_KEY=
# Needed only for MING_VOICE_PROVIDER=deepgram.
DEEPGRAM_API_KEY=
# DEEPGRAM_STT_MODEL=nova-3
# Who transcribes you in deepgram mode: deepgram (default) or openai. OpenAI
# STT uses OPENAI_API_KEY and falls back to Deepgram when it fails; Firefox's
# ogg recordings always go to Deepgram. DEEPGRAM_API_KEY stays required.
# MING_STT_PROVIDER=deepgram
# OPENAI_STT_MODEL=gpt-transcribe
# DEEPGRAM_TTS_MODEL=aura-2-orion-en
# Bart's spoken voice in deepgram mode: deepgram (default), gemini or openai.
# Gemini TTS uses the Gemini key (precedence below); OpenAI TTS uses
# OPENAI_API_KEY. Either falls back to Deepgram when it fails before any audio.
# MING_TTS_PROVIDER=deepgram
# GEMINI_TTS_MODEL=gemini-3.8-flash-lite-tts
# GEMINI_TTS_VOICE=Puck
# OpenAI shuts down its speech endpoint on 2027-01-06; Deepgram answers after that.
# OPENAI_TTS_MODEL=gpt-4o-mini-tts
# OPENAI_TTS_VOICE=cedar

# Live avatar: none (default, looping call video) or simli (Bart lip-synced to
# the OpenAI Live voice). Needs MING_VOICE_PROVIDER=openai. See docs/LIVE_AVATAR.md.
MING_AVATAR_PROVIDER=none
# Needed only for MING_AVATAR_PROVIDER=simli. Both stay server-side.
SIMLI_API_KEY=
SIMLI_FACE_ID=
# Optional: livekit (default, relayed, best on restrictive Wi-Fi) or p2p
# (direct WebRTC, slightly lower latency). Sessions end after this many minutes.
# MING_AVATAR_TRANSPORT=livekit
# MING_AVATAR_MAX_MINUTES=20
# Framing, per face (Trinity-2 faces only): zoom below 1 shows the shoulders;
# shift moves the avatar right in half video widths to centre an off-centre face.
# MING_AVATAR_ZOOM=0.65
# MING_AVATAR_SHIFT_X=0.25

# Gemini/Veo key precedence: GEMINI_API_KEY, GOOGLE_GENERATIVE_AI_API_KEY,
# then GOOGLE_API_KEY. Only the first nonblank key is used.
# GOOGLE_API_KEY=
# Keys stay server-side. No key is needed just to open the local interface.

# Optional advanced settings. npm run dev generates an internal gateway token
# automatically when none is configured; there is no presenter login.
# MING_BACKEND_TOKEN=
# MING_BACKEND_URL=http://127.0.0.1:4111
# Usually leave unset so localhost and 127.0.0.1 both work.
# MING_APP_ORIGIN=http://127.0.0.1:3000
# Container mode needs a fixed shared token and persistent database volume.
# MING_DATABASE_URL=file:/data/storage.db