Download

Run a node on Silicon Network.

The engine serves open models on an Apple Silicon Mac; the node agent offers a Linux GPU rig or server for rendering, training and mining. Both are Apache-2.0, ship as source, and join with your operator account.

The first release is on its way

No engine release is published yet. The files below appear here as soon as one is, and the commands on this page fetch the newest one on their own.

Engine for Mac

Apple Silicon with macOS 14 or newer. One line checks the Mac, verifies the signed release with the key below, installs it and starts the engine at login. Then join from the engine dashboard with device sign-in, as the operator guide describes.

silicon - zsh

curl -fsSL https://www.siliconnetwork.io/install.sh | sh

Read the script first if you like: it is the whole installer. A source release builds on the Mac, so it asks for the Xcode command line tools, Rust and Apple's Metal toolchain when they are missing.

silicon - zsh

# see every check and download without changing anythingcurl -fsSL https://www.siliconnetwork.io/install.sh | sh -s -- --dry-run # remove the engine and its launchd agents (keeps models and engine.env)curl -fsSL https://www.siliconnetwork.io/install.sh | sh -s -- --uninstall

Or step by step

The same release by hand, with uv and Rust installed.

silicon - zsh

# the newest engine release, checked against its SHA256SUMSV=$(curl -fsS https://www.siliconnetwork.io/releases/latest/version)curl -fsSLO https://www.siliconnetwork.io/releases/v$V/silicon-engine-$V-source.tar.gzcurl -fsSLO https://www.siliconnetwork.io/releases/v$V/SHA256SUMSgrep " silicon-engine-$V-source.tar.gz$" SHA256SUMS | shasum -a 256 -ctar -xzf silicon-engine-$V-source.tar.gz && cd silicon-engine-$V # once, on macOSxcodebuild -downloadComponent MetalToolchain # run it; --extra mlx is required on a Mac (the dashboard comes built)SILICON_API_KEY=choose-a-long-secret uv run --extra mlx silicon

What a Mac could earn

Pick your Mac, how long it serves and what you pay for power. The estimate uses the network's published prices and operator share; its assumptions are listed under it. Running several Macs? The fleet guide shows the cost per token and payback math.

Estimate, not a promise

MacMemoryHours a day it servesBusy share of those hours (%)Electricity, USD per kWhDraw while generating (W)

Best fit: Llama 3.2 1B; about 5.7M tokens a day
Credit earned, a month: $178.16
Power cost, a month: $2.14; 12.6 kWh

Models a Mac mini (M4 Pro) with 48 GB can hold, with estimated speed, tokens, credit and power a day
Model it can serve	Tokens per second	Tokens a day	Credit a day	After power
Llama 3.2 1B *	262 est.	5.7M	$5.94	$5.87
Llama 3.2 3B *	103 est.	2.2M	$2.33	$2.25
GPT-OSS 20B *	93 est.	2.0M	$2.10	$2.03
Qwen3-VL 4B *	57 est.	1.2M	$1.30	$1.23
Llama 3.1 8B *	41 est.	890K	$0.93	$0.86

* No rate of its own on the price list: it bills at the network default.

Too large for one Mac with 48 GB: Llama 3.1 70B (41 GB of weights), Llama 3.3 70B (41 GB of weights), DeepSeek V3.1 (406 GB of weights), DeepSeek V3.2 (378 GB of weights), Qwen3 235B (142 GB of weights), Kimi K2 Thinking (707 GB of weights), GLM 4.7 (199 GB of weights), GPT-OSS 120B (71 GB of weights).

The assumptions

Operators earn 70% of what customers pay for the work their Macs serve, the share in the network's code (the published figure could not be read just now). Credit accrues today; payouts are not switched on yet.
The price list could not be read, so every model is priced at the network default in the code: $1.50 per 1M output tokens. Prompt tokens also earn credit and are left out, so the figures lean low there.
Speed is the network's measured median for this chip and model where there is one. Otherwise it is estimated: generating a token reads the model's active weights once, so tokens per second is about 70% of the memory bandwidth (273 GB/s, Apple's published figure for this Mac) over the bytes read per token. Long prompts and many parallel requests change it both ways.
A model fits when its weights plus 15% for the cache fit in the memory macOS lets the GPU use (two thirds below 36 GB, three quarters from 36 GB). Larger models can be split across several Macs; this estimate covers one.
The busy share is how much of the serving time the Mac is generating. It depends on demand, which is not in your control and is low while the network is new. The rest of the serving hours the Mac waits at about 5 W.
The draw while generating (65 W) is our assumption for a Mac mini (M4 Pro), below its rated maximum. A plug meter tells you yours; enter it above.
A month is 30 days. Wear, internet, taxes and the Mac itself are not included.

Node agent for Linux

Linux on x86_64 or arm64, Python 3.11 or newer, and Podman or Docker. The Linux agent guide covers the config, prices and the systemd unit.

root - bash

# as root, on the Linux machine: the newest release, checked against its SHA256SUMSV=$(curl -fsS https://www.siliconnetwork.io/releases/latest/version)cd /tmpcurl -fsSLO https://www.siliconnetwork.io/releases/v$V/silicon-node-agent-$V.tar.gzcurl -fsSLO https://www.siliconnetwork.io/releases/v$V/SHA256SUMSgrep " silicon-node-agent-$V.tar.gz$" SHA256SUMS | sha256sum -cinstall -d -m 755 /opt/silicon-node-agenttar -xzf silicon-node-agent-$V.tar.gz -C /opt/silicon-node-agentpython3 -m venv /opt/silicon-node-agent/.venv/opt/silicon-node-agent/.venv/bin/pip install --require-hashes \  -r /opt/silicon-node-agent/node-agent/requirements.txt # the credential, readable by root onlyinstall -d -m 700 /etc/silicon-node-agentsh -c 'umask 077; printf "%s\n" "snop_..." > /etc/silicon-node-agent/credential' # the config: accepts nothing until you list workloads and pricescp /opt/silicon-node-agent/node-agent/agent.example.toml /etc/silicon-node-agent/agent.toml # check the config and the hardware (sends nothing)cd /opt/silicon-node-agent/node-agent/opt/silicon-node-agent/.venv/bin/python -m silicon_node_agent check/opt/silicon-node-agent/.venv/bin/python -m silicon_node_agent detect

Check what you download

Every file is listed with its SHA-256 in SHA256SUMS, and the commands above stop if a file does not match. manifest.json carries the same checksums with an Ed25519 signature from the release key below; the engine's updater accepts an update only with that signature, from a key pinned in its source.

Key id: release-2026-10
Public key: HKn1m5b5rXNF95yzQ4Qs9QuJCHjqvPIlstK0YRHGDp4=