Integrations

AI providers

Jarvis talks to a pool of providers behind the scenes. You don't choose one - the bot picks the best available response for each turn.

The pool currently includes NVIDIA NIM, Mistral, Google Gemini, Ollama Cloud, OpenAI, and DeepSeek. Models and availability change over time. Requests sent to an external provider are processed under that provider's terms; locally hosted services remain on the Jarvis host.

Voice stack - NVIDIA NIM + Fish Audio

Voice is its own pipeline. When you run /voice, Jarvis joins the channel and starts streaming audio chunks through:

1. Wake-word detection - local model listens for "jarvis" / "garmin" / your custom word
2. Speech-to-text - NVIDIA NIM Parakeet, post wake-word audio only
3. AI response - same provider pool used for text chat
4. Text-to-speech - Fish Audio when configured, otherwise NVIDIA NIM TTS, streamed back into the voice channel

Raw audio is never persisted. Transcripts of what is actually spoken are what leaves the box - in a solo call Jarvis may pick up speech without a wake word for lower latency, so the wake word is not a guarantee in every mode (see the Privacy page), and those follow the same 30-day memory retention as text chat.

Self-hosters: NIM access is optional. The bot runs without voice if you don't configure NIM credentials - it just won't respond to /voice.

Music sources

• YouTube - searches, direct URLs, and playlists resolve through the local Lavalink node and begin streaming without a download step.
• SoundCloud - public searches and track URLs resolve through the same local Lavalink node. Private or login-walled tracks are not available.
• Direct file upload - attach up to 9 audio files (30 MB each) to one /play call. MP3, WAV, OGG, FLAC, M4A, and OPUS are accepted.

Uploaded codecs may be normalized with ffmpeg; YouTube and SoundCloud playback stays inside the local Lavalink pipeline.

Storage & portal

Per-server config, blacklists, warning logs, and encrypted memory records live in the local SQLite store. Memory payloads are encrypted with the configured master key.

The web portal uses Discord OAuth for sign-in. Sessions are HTTP-only cookies. Once signed in, you can manage automod lists, role/channel restrictions, and feature toggles from a UI instead of slash commands.

Webhooks & API

Jarvis exposes a small JSON API on /api/stats (public guild/user counts and 24h request volume) and an internal webhook receiver. The webhook endpoint signs incoming bodies and is intended for internal release automation, not third-party integrations.

Endpoints worth knowing:

• GET /api/stats - public, returns { guildCount, userCount, requests24h }
• GET /health - public liveness, returns 503 when the database is unreachable. Add ?deep=1 with a bearer HEALTH_TOKEN for the verbose provider and environment snapshot.
• GET /metrics/commands - bearer-gated command usage breakdown
• GET /sitemap.xml, GET /robots.txt - SEO surface