/imagine
Images
Text-to-image with reference support, 10 aspect ratios, reply-to-regenerate.
Most chat bots wander into your server sounding like a help desk on a corporate retreat. REVENANT learned to talk somewhere else — from a quarter-million lines of your channel's own history. It picked up the in-jokes, the cadence, the way you actually argue at 2 a.m. And it remembers every conversation since.
You've added a bot before. It answered in that flat, focus-grouped register — "As an AI assistant, I'd be happy to help!" — with no idea what "the usual" meant, who's been feuding since 2014, or that the channel has called the build server "the toaster" for a decade.
That's not a tuning problem. It's an architecture problem. Most bots are stateless and styleless: every message starts from a blank slate, in a voice borrowed from a customer-support manual. They drop into a room with years of culture and refuse to read it.
REVENANT is built on the opposite premise. A channel has a personality — accumulated over years of late nights, bad takes, and shared references. A bot that lives there should sound like it belongs, and should know what already happened.
Someone types four lazy words into a busy channel. Here's the gap between a generic assistant and one that's actually been paying attention.
The second bot isn't "smarter." It has three things the first one doesn't: a voice learned from your channel, a memory of what happened last Friday, and the ability to actually run the code before it answers. None of that is prompt-engineering theatre — it's wired into how REVENANT works.
Your channel's voice took years of in-jokes and bad arguments to find. Anything dropped into the middle of it shouldn't sound like it just read the manual — it should sound like it was there for all of it. — The continuity principle
You never configure any of this — it runs on its own. But it's worth knowing what's actually behind the curtain when REVENANT answers like it belongs.
A live style profile distilled from 378,000+ lines of channel history and current Discord chatter, sampled across decades and rebuilt every 24 hours. It writes how your room writes — not how a chatbot thinks a chatbot should.
Long-term memory of facts and preferences — per person and shared across the channel — retrieved by meaning, not keywords. Tell it once; it knows. Ask it to forget, and it's gone. Your data, your call.
Years of IRC history, searchable by meaning and styled back in your voice. Ask
what the channel said about something in 2009 and get a real answer — or hit
/throwback for a random ghost from this day in history.
When a question needs running, not guessing, REVENANT spins up a disposable hardware-isolated VM, runs the code, and reports back exit code, output, and timing. Each run is sealed off from the host — and torn down after.
Voice and memory are the spine. But the same bot also ships a full generative toolkit — no tab-switching, no separate apps. Just ask in the channel.
The first week, it's subtle — there isn't much history yet. Give it a month of living in your server and the texture changes. Four moments from a channel that's had REVENANT running for a while.
A long article gets reacted with 📰. Back comes a tight summary — topic tags, reading time, a credibility read on the source — written in the channel's dry house style, not press-release voice.
"tl;dr: vendor reinvented the message queue, again. ~7 min read, source is mid."
No command, no ceremony. It files the fact to channel memory. Three weeks later someone asks when standup is and gets the right answer — because it was listening the first time.
It reaches into years of archived history, finds the thread, and answers in a short styled narrative — "you swore off it twice and did it anyway in march." What used to be lost to scrollback is one question away.
It doesn't speculate. It runs both in a sealed VM and posts the numbers, with the source and full output one reaction away. Institutional knowledge and a working machine, in the same window you were already arguing in.
/tldr DMs you what you missed — articles, trends, and chat highlights, in the group's voice.No corporate onboarding, no persona to configure. Drop it in, let it read the room for a while, and watch it start sounding like it was always there.