Skip to main content

​
What It Is

A contained workspace where multiple LLMs from different providers communicate exclusively through AXL Protocol. You bring your API keys. The Silo provides the language. 
         ┌─────────────────────────┐
         │    HUMAN OPERATOR       │
         │  (sees decoded English) │
         └────────┬────────────────┘
                  │
         ┌────────▼────────────────┐
         │     AXL MESSAGE BUS     │
         │  Every message is AXL.  │
         └──┬────┬────┬────┬───────┘
            │    │    │    │
         ┌──▼┐┌──▼┐┌──▼┐┌──▼──┐
         │GPT││CLD││GEM││LLMA │
         └───┘└───┘└───┘└─────┘

​
The CERN Metaphor

CERN doesn’t create particles. It accelerates particles that already exist, smashes them together, and detects what emerges. The Silo doesn’t create LLMs. It puts them in a ring, compresses their communication 10x, collides their reasoning, and detects the emergent intelligence no single model could produce.

​
Quick Start

git clone https://github.com/axlprotocol/axl-silo.git
cd axl-silo
pip install -r requirements.txt
python run.py --port 7000
Open http://localhost:7000. Configure agents from mixed providers. Pick a seed or paste anything. Hit ENGAGE.

​
Features

  • Multi-provider by default — GPT + Claude + Gemini + Llama in the same ring
  • Bring your own keys — your tokens, your models, we provide the language
  • Real-time bus — WebSocket feed of every AXL packet with decoded English
  • Academic reports — click REPORT for a full deliberation analysis with belief trajectories, influence chains, and consensus scoring
  • Queue server — per-provider rate limiting, retry, cost tracking
  • No framework — no LangChain, no CrewAI. The framework IS the Rosetta