What is Jarvis? A free Discord AI bot, explained

Jarvis is a Discord bot that talks back. It joins your server, listens when you mention it, replies in chat or out loud in voice channels, remembers what you've talked about, and ships with the usual moderation toolkit on the side. Free to use, no premium tier.

That's the short version. Here's the longer one.

Where it came from

Jarvis is the public bot built by AGIS - Artificial General Intelligent System, the Discord server where the project lives. It started as a single-server experiment in late 2024 and has been running publicly since. The "Jarvis" name is a nod to the obvious; the AGIS domain runs at jorvis.org, hence the alternate spelling that shows up in some URLs.

The bot exists because most Discord AI options at the time were either paywalled, locked to one model, or didn't handle voice and memory together. Jarvis bundles all of that into one bot you don't pay for.

What it actually does

Five capability buckets, all in one bot:

Chat. Mention @Jarvis, run /jarvis prompt: …, reply to one of his messages, or say his name as a wake word. He answers using a rotating pool of AI providers - OpenRouter, Mistral, Google Gemini, Groq, Cerebras, DeepSeek - whichever responds fastest that turn. There's no single hard-coded model.

Voice. Run /voice while in a voice channel and Jarvis joins, listens for the wake word, transcribes, generates a response, and speaks it back. Voice runs on NVIDIA NIM - Parakeet for speech-to-text and a NIM model for TTS. Raw audio is never persisted.

Memory. Conversations carry over between sessions. Memory is per-user and tied to your account - it follows you across every server you share with Jarvis, not locked to one, and removing the bot from a server does not delete it. Default retention is 30 days; /clear wipes it instantly. /opt mode: out disables memory storage entirely and erases it everywhere.

Music. /play streams from YouTube, SoundCloud, or attached audio files. Per-server queue, loop modes, optional DJ-only restriction.

Moderation. Standard moderator kit - /automod, /purge, /timeout, /ban, /kick, /warn, /role, /channel. Every command checks the Discord permission it would normally require; Jarvis doesn't override Discord's permission model.

What makes it different

A few design choices that aren't standard for the category:

• Multi-provider routing. Most AI bots lock you to one model. Jarvis rotates the pool per request, so a single provider's outage or rate-limit doesn't take the bot down.
• No premium tier. No monthly subscription, no message cap, no model gating. The full feature set is free, full stop.
• Voice is native, not a plugin. STT, response, and TTS run as one continuous pipeline. There's no separate "voice bot" component.
• Memory you can audit. /memory entries: 10 shows you what he's stored about you. Most bots either don't have memory or don't let you see it.
• Actively maintained. Releases ship continuously; bug reports filed in the support server often turn into same-week fixes.

What it isn't

Worth being clear about. Jarvis is not:

• A general-purpose AI assistant outside Discord. He lives in Discord. There's no API, mobile app, or web dashboard for chatting with him directly.
• A roleplay bot. He'll chat in character if asked, but he's not optimized for long-form persona scenes the way Character.AI or NovelAI clones are.
• A "smart" replacement for human moderators. AutoMod handles keyword filtering; it doesn't make judgment calls about context or intent.
• Always on the latest model. The router optimizes for latency and availability, not benchmark scores. Frontier models cycle in when they're fast enough.

Who it's for

Servers that want one bot to do AI chat, voice replies, music, and moderation without paying for premium and without wiring four different bots together. Mid-sized communities, friend groups, coding/dev Discords, study groups. Anywhere someone would type "hey, can someone summarize that?" and want an answer in five seconds.

Try it

Invite Jarvis to a server you manage, or browse the docs first. Setup is three steps and takes about thirty seconds - see Getting started for the walkthrough.