Skip to content

01 — What is The Aetherium Engine?

The Aetherium Engine is a multi-agent AI orchestration platform that runs locally. It coordinates teams of AI agents — supervisors, workers, specialists — so they can collaborate on complex tasks instead of you driving a single chat session.

Who this is for

The Engine is built for developers and development teams who want multi-agent orchestration on their own hardware. If you've spent time wiring open-source agent frameworks into a production workflow and bounced off the surface area, the Engine is what you wanted those tools to be: a complete platform with a UI, a database, an inference router, a tool execution sandbox, and a real-time chat interface — not a kit you have to assemble.

How an agent team works

A user message arrives. A supervisor agent plans. Workers execute, calling tools to read files, run shells, search the web, query knowledge bases, generate images. Workers report back. The supervisor synthesizes a response. Every step streams to the browser as it happens.

graph TB
    User([User])
    Sup[Supervisor]
    W1[Frontend Worker]
    W2[Backend Worker]
    T1[(workspace tools)]
    T2[(quality tools)]

    User -->|"Build me a login page"| Sup
    Sup -->|delegate| W1
    Sup -->|delegate| W2
    W1 -.->|"write_file login.tsx"| T1
    W2 -.->|"run_tests"| T2
    W1 -->|reply| Sup
    W2 -->|reply| Sup
    Sup -->|synthesized response| User

The example above is Hierarchical topology. The Engine ships with four:

Topology Pattern Use case
Hierarchical Supervisor delegates to workers Complex tasks with clear delegation paths
Sequential Pipeline where each agent builds on the previous output Document processing, multi-stage analysis
Roundtable Agents respond in parallel, then critique and reconcile over discussion rounds Brainstorming, diverse perspectives
Single One agent responds directly Simple queries, single-domain tasks

What's in the box

The Engine ships with 12 MCP server classes exposing 44 tools — file operations, shell execution, web research, knowledge base CRUD, code quality and review, planning, image and video generation, voice-over, inter-agent communication, and a VR headset. Agents are wired together by Workflow Templates (one-click project creation) or generated from natural-language descriptions via Quick Create.

The flagship reference deployment is an 8-agent Steam Frame VR development pipeline, shipped as a template. The Engine itself is general-purpose — Steam Frame VR is one example of what an 8-agent Hierarchical team can do, not the Engine's purpose.

Why local

  • Data sovereignty. Prompts, code, tool outputs, and conversation history never leave your network. Knowledge bases live on your disk.
  • Cost predictability. No per-token billing. The hardware is the only ongoing cost.
  • No vendor lock-in. The Engine talks to any OpenAI-compatible inference server. LM Studio is the default; anything that speaks the same API works.

Where to go next