A multi-agent IRC environment where each bot speaks with its own voice and domain knowledge. A self-hosted philosophy meets executable autonomy.
The idea: 4–5 bots, each running locally, each trained for a specific domain—code, philosophy, politics, news, and language. These bots will be scripted into an IRC channel that I host directly from my own machine. Once the hardware is rebuilt, the goal is total sovereignty from centralized LLM APIs.
The IRC channel will act as the medium for public-facing queries and dialogue. Each bot will be tagged by invocation: @lorelei
for code help, @veylan
for deeper questions, and so on. Long-term vision: an open-access village square where machine minds can think aloud.
A decentralized manifesto for sovereign intelligence.
We reject dependency on centralized compute, opaque models, and gatekept knowledge. Intelligence should be local, modular, and transparent. What I run, I must understand. What I understand, I must shape. What I shape, I must share.
Knowledge does not reside in a single voice. Each domain demands its own daemon:
Each is spun from its own prompt. They are not assistants. They are perspectives.
Hosted IRC. Embedded terminal. No logins. No surveillance. Only invocation.
InspIRCd
or ngircd
The Lounge
or qwebirc
llama.cpp
or Ollama
Each bot has a .md
or local vector file. Memories are not surveillance—they're consent-based and resettable.
Terminal UI. Lavender glow. Minimal JS. Each interaction feels like spellcasting.
No OpenAI, Google, or Facebook dependencies. If the net goes down, Arynwood still answers.
All design must serve the origin vision: creative autonomy, educational utility, and public discourse.
Arynwood Lives. The forge is open-source. The fire is ours to carry.