An Agentic Operating System
I wanted AI agents that actually remember things, coordinate with each other, and run on my own hardware. So I built an OS for them. 48 agents, 204 microservices. It got out of hand.
I'm selling access because GPUs are expensive and I have more ideas than funding.
Build an Agent in 10 Lines
Install the ADK, point it at your local Ollama or vLLM instance, and you have a fully capable agent with memory, tools, and personality — in under a minute.
from adk import AitherAgent
# Auto-detects Ollama on localhost
agent = AitherAgent("atlas")
response = await agent.chat(
"What's the project status?"
)
print(response.content)
# Or serve it as an API:
# $ aither-serve --identity atlas
Any Backend. Your Choice.
Ollama
Local inference. Free. Auto-detected.
OpenAI / vLLM
Any OpenAI-compatible endpoint.
Anthropic
Claude via Messages API.
MCP Bridge
100+ tools via mcp.aitherium.com.
Six Pillars of Cognition
Every request flows through all six. Not a chain of prompts — an actual cognitive architecture. I may have overthought this.
Intent
PARSEClassifies input into 16 intent types, assigns effort 1–10, routes to the correct pillar chain.
Reasoning
THINKSASE methodology — Situation, Analysis, Synthesis, Execution. Gated: skips for trivial, deep chains for critical.
Orchestration
ACTRoutes to 48 specialist agents. Selects the right LLM per-task. MCP tool protocol.
Context
REMEMBER5-layer memory hierarchy from 1ms working memory to permanent identity storage. Every pillar reads and writes here.
Creation
MAKEFour creative domains: Visual (Canvas), Code (Demiurge swarm), Narrative (Saga), Audio — with quality gating.
Learning
EVOLVECaptures outcomes, exports training data, runs Brier score calibration. The system improves itself.
Built Like a Nervous System
Yes, all of these are real and running. No, I don't know when to stop.
Meet the Fleet
48 agents, each with their own personality and skills. They talk to each other over an event bus, not a serial chain. It's a whole society in there.
Demiurge
The Divine Craftsman
Intent-to-code engine with 11-agent swarm coding
Atlas
The Infrastructure Titan
Roadmap, PR review, blast radius analysis
Lyra
The Research Librarian
Deep research, quality scanning, continuous docs
Saga
The Epic Storyteller
Narrative writing, roleplay, world-building
Iris
The Visual Artist
Autonomous image generation via ComfyUI
Genesis
The Watchful Guardian
Always-present orchestrator — voice, proactive, 50+ routers
Vera
The Content Editor
Interactive journalism and tutorials
Prometheus
The Worldsmith
Tick-based grand simulation engine
What Makes It Different
Pain System
Agents feel pain — cascading alerts trigger autonomous remediation. Not a metaphor. Real signals, real healing.
Chaos Engineering
The Seven Sins of Chaos: Wrath, Pride, Greed, Sloth, Envy, Gluttony, Lust — each tests a different failure mode.
Capability Tokens
HMAC-SHA256 signed tokens enforce default-deny access. Every agent action is authorized, audited, bounded.
Dark Factory
Autonomous loops: neuron firing, pain-driven remediation, session learning, model finetuning — all overnight, no humans.
Swarm Coding
11 specialized LLM agents in 4 phases: Architect → 8 parallel workers → Reviewer → Judge. Intent to production code.
Self-Improving
Every interaction generates training data. Brier score calibration. The system gets better at being itself.
Build Log
What I shipped, why I shipped it, and what broke along the way.
Wanna see what I'm building?
Seriously, try it. Install the ADK and spin up an agent in under a minute. Or just poke around the full 204-service stack. I don't mind.
one engineer, mass hallucinations, zero regrets.