Reachy Mini App

Ante

Ask it anything, out loud. The model doing the thinking runs on the 0G Compute Network — a decentralised inference network, not one company's data centre.

Read this first

You need your own 0G account

Ante has no hosted service behind it and no shared API key. Its hearing and its thinking are paid for per request from your own 0G wallet, so installing the app is not enough by itself. Without a funded account, Ante will hear you and then decline to answer with insufficient balance.

npm i -g @0glabs/0g-serving-broker
0g-compute-cli login
0g-compute-cli deposit --amount 4

# a funded sub-account for each of the two providers
0g-compute-cli transfer-fund --service inference --amount 2 \
  --provider 0x7DCFe6AEa70350C2090041524c9B4A9262DCe87D   # thinking: glm-5.2
0g-compute-cli transfer-fund --service inference --amount 2 \
  --provider 0x36aCffCEa3CCe07cAdd1740Ad992dB16Ab324517   # hearing: whisper-large-v3

Fund each provider with headroom rather than the bare minimum. The required balance is a locked reserve plus unsettled fees, and the reserve cannot be spent from, so funding exactly the minimum buys a single request.

Getting Started

Installation

Install from the Reachy Mini app store, or run it locally. Ante needs two processes: the 0G speech backend, and the app itself.

uv sync
cp .env.example .env

ante-0g-server            # add --push-to-talk for the antenna trigger
ante                      # in a second terminal

Speech synthesis runs locally, because the 0G network serves chat, transcription, image and video models but no speech synthesis. On the Wireless robot install Piper and point ANTE_PIPER_VOICE at a voice file; on a Mac the server falls back to the system voice for development.

Before you install

What it feels like

A turn takes about thirteen seconds from the moment you stop speaking to the moment Ante starts: roughly five seconds to transcribe and eight to reply. That is the network's round-trip rather than thinking time, and the two calls cannot overlap, since there is nothing to think about until the audio has been transcribed. Ante is written to answer briefly partly because of this.

That is the trade. Nobody can read your conversations out of a provider's logs, and no company can swap the model out from under you. In exchange you keep a wallet topped up and you wait a few seconds longer.