Inference

Models for agents
that move money.

Agents ask models in a loop, on data they cannot leak. On Amadeus they run sealed with the leading models today, and our own miners already do the maths those models run on.

1.5T+
Matrix multiplications a day
20M
Every second, around the clock
0.5s
Finality
Mar 2025
Live since

WORKS WITH

  • Uniswap
  • Jupiter
  • Robinhood
  • Solana
  • Ethereum
  • Base
  • Sui
  • Hyperliquid
  • Lighter
  • Treble Finance
  • Tenstorrent
  • Arweave
  • Aarc
  • Neofin
  • Legal Nodes
  • CertiK
  • CoinGecko
  • CoW Swap
  • LI.FI
  • Velora (formerly ParaSwap)
  • OpenOcean
  • KyberSwap
  • SushiSwap
  • NEAR
  • Ostium
  • Binance
01 — Why it matters

Agents need a different kind of inference

Four reasons a public AI service is the wrong place for an agent that moves money.

01
01

Agents ask in a loop

A person asks a model once. An agent asks thousands of times a day, and every call adds cost and delay.

With Amadeus: Agents on Amadeus come with the leading models built in, and our own miners will serve them next.

02
02

The data cannot leave

Positions, strategies and client data are too sensitive to send to a public AI service.

With Amadeus: The agent and its data run inside a sealed machine. Miner-served inference brings the model inside the same seal.

03
03

Answers cannot be replayed

When an AI decision moves money, you need to show what it saw and why it acted.

With Amadeus: Same inputs, same output, with a receipt you can replay months later.

04
04

Mining that serves no one

Most proof-of-work burns energy on maths nobody needed.

With Amadeus: Amadeus miners do the matrix maths that models run on, and the network checks every result.

02 — How it works

Inference on our own network

The network already does the maths models run on. Serving agents is the job we give it.

01
01

Verified compute

Miners do the matrix multiplications models are built on. Other nodes check every result.

02
02

Leading models for agents

Agents on AMA Hub and AMA Embed use the leading models today, from inside their sealed machines.

03
03

Served by our miners

Next: the same models served by uPoW miners, so inference runs on the network that already does its maths.

04
04

Nova on uPoW

Later: Amadeus’s own model, trained and served by the same miners.

Private inference for your agents.

Tell us which models your agents call. We show you how they run on Amadeus today, and on our own miners next.