//POE · SHA-256 COMMITMENTNODES REGISTERED13//COORDINATION · ETHEREUM L1STAKED1,898,598 $PRLX//NO SLASHING · PRINCIPAL RETURNED IN FULLREWARD POOL502,878 $PRLX//PARALLEL-NATIVE · REQUEST ROUTINGSTREAMING0.4 $PRLX/S//$PRLX · 100M FIXED · 0% INFLATIONREWARDSPER-SECOND · UPTIME-WEIGHTED//PARALLELIX AI · OPEN-SOURCE INFERENCE//POE · SHA-256 COMMITMENTNODES REGISTERED13//COORDINATION · ETHEREUM L1STAKED1,898,598 $PRLX//NO SLASHING · PRINCIPAL RETURNED IN FULLREWARD POOL502,878 $PRLX//PARALLEL-NATIVE · REQUEST ROUTINGSTREAMING0.4 $PRLX/S//$PRLX · 100M FIXED · 0% INFLATIONREWARDSPER-SECOND · UPTIME-WEIGHTED//PARALLELIX AI · OPEN-SOURCE INFERENCE
network
ParalleliX.
OPERATOR HANDBOOK

RUN A NODE.EARN FOR UPTIME AND AI JOBS.

Single binary. One setup command. Run parallelix-node setup and it detects your hardware, checks Ollama, recommends models sized to your VRAM, generates a node key, prints a prefilled staking link, then auto-detects your node on-chain the moment the stake lands. After that: start the daemon and earn for uptime.

v2.0.0
$curl-fsSLhttps://parallelix.io/install.sh|sh

Or read the operator manual

linux x86_64available
macOS arm64available
windows · wsl2available
  • v2.0.0
  • x86_64 + arm64
  • secp256k1 signed

Windows runs under WSL2. Native Windows is not supported; install the CLI inside a WSL2 Linux distribution.

WHY RUN A NODE

THREE OPERATIONAL FACTS.

What the node does, what the demand looks like, and how uptime becomes $PRLX. Read each card top to bottom.

  • // operator

    IDLE HARDWARE EARNS.

    Most GPUs spend most of their hours waiting. Keep a node online and rewards stream to it per-second, weighted by stake, tier, and uptime.

    Specs
  • // scheduler

    DEMAND: AI INFERENCE.

    Demand comes from ParalleliX AI, running open-source LLM inference on network compute. Your node receives whole inference requests routed to its hardware tier.

    Lanes
  • // ledger

    UPTIME ACCRUES. CLAIM ANYTIME.

    No per-task invoices. Your uptime sets your weight (stake x tier x uptime) and the contract streams your slice of the rewards per-second while your node is online. Accrued $PRLX is claimable at any time via the Console.

    Mechanics
THE BINARY

ONE COMMAND, SIX SUBCOMMANDS.

parallelix-node is a single binary. setup runs the full onboarding sequence and auto-detects your node on-chain. start and service skip --node-id once setup has saved it. The on-chain half is registerNode() in the Console, reached via the prefilled link setup prints. Terminal output below is illustrative.

  • // 01

    SETUP

    One-command operator onboarding. Detects hardware tier, checks Ollama, recommends and pulls models sized to your VRAM, generates the node key (nodeKeyHash), prints a prefilled staking link, then polls the chain and saves your node id the moment the stake lands. Offers to install the 24/7 service when done.

  • // 02

    STAKE

    On-chain half via the Console. setup prints a prefilled link that arrives with keyHash and tier already filled. Your staking wallet calls registerNode() on NodeRegistryLocker. Stake 50,000 $PRLX or more per node. No slashing. 7-day unstake cooldown.

  • // 03

    START

    --node-id is optional once setup has saved it. Daemon mode: liveness heartbeats for uptime, encrypted inference requests dispatched by the coordinator, each request runs sandboxed (CUDA on GPU nodes, CPU fallback), result bytes returned with a PoE commitment. Live dashboard shows status, lifetime earnings, and per-request SHA-256 ticks.

  • // 04

    PROOF

    Every result ships a SHA-256 commitment over (request_id, result, node_id), anchored on-chain. A hash commitment, not a zk-SNARK. verify re-checks any stored proof locally.

  • // 05

    MODELS

    catalog lists the full supported model set. recommend filters it against your detected VRAM: Mistral 7B, Qwen 2.5 (7B/14B/32B), DeepSeek-R1 8B, Gemma 2 9B, Phi-4, Llama 3.2/3.3 70B. Any Ollama model also works.

  • // 06

    UPDATE

    Self-update: fetches the latest version, verifies sha256 against parallelix.io/cli/manifest.json, swaps in the new binary atomically with a backup. Under systemd the node logs clean timestamped dispatch, PoE, and uptime-credit lines; no per-task settlement line since uptime accrual is the reward mechanism.

LIFECYCLE

INSTALL TO ACCRUE.

One operator session, end to end. Five stages. Install the binary, run setup (hardware detect, model pull, node key, prefilled staking link, on-chain auto-detect), stake in the Console, start the daemon, and accrue streaming uptime rewards.

  1. // 01

    INSTALL

    One command, zero prerequisites. On Linux the script bootstraps the whole operator stack: Node.js, Ollama (the local GPU runtime), and parallelix-node in /usr/local/bin. macOS installs Node plus the CLI and points you at the Ollama app. The installer ends by offering to run setup immediately.

    parallelix-node // install
    $ curl -fsSL https://parallelix.io/install.sh | sh
    > installing Node.js 22 LTS ok
    > installing Ollama (GPU runtime) ok
    > installed /usr/local/bin/parallelix-node
  2. // 02

    SETUP

    Run parallelix-node setup once. It detects your hardware tier [1/5], checks Ollama [2/5], recommends and pulls a model catalog sized to your VRAM [3/5], generates the node key (a low-value secp256k1 keypair the CLI uses to sign liveness and results, never your staking wallet) [4/5], then prints a prefilled staking link with keyHash and tier already filled [5/5].

    parallelix-node // setup
    $ parallelix-node setup
    > [1/5] cpu: ryzen 9 7950x · gpu: rtx 4090 24gb · tier t2
    > [2/5] ollama 0.3.9 · running
    > [3/5] pulling qwen2.5:14b
    > [4/5] node key · secp256k1 0xa1f3...c7b4 saved
    > [5/5] stake at app.parallelix.io/register?keyHash=0xa1f3...&tier=t2
  3. // 03

    STAKE

    Open the prefilled link from setup. Your staking wallet calls registerNode() on NodeRegistryLocker: stake 50,000 $PRLX or more per node, confirm the tier, submit. No slashing: principal is always returned in full after the 7-day unstake cooldown. setup polls the chain and auto-detects your node id the moment the transaction lands, then saves it locally.

    parallelix-node // stake
    // open the prefilled link · connect wallet · confirm registerNode()
    > contract: NodeRegistryLocker · stake >= 50000 $PRLX
    > polling chain...
    > node 7 detected on-chain · node id saved
  4. // 04

    START

    --node-id is optional once setup has saved it. The daemon heartbeats for uptime and accepts inference requests the coordinator dispatches from ParalleliX AI. The sandboxed runtime decrypts each request, runs it whole (CUDA on GPU nodes, CPU fallback on edge), and returns the result bytes with a PoE commitment. The live terminal dashboard shows status, lifetime earnings, and per-request PoE ticks.

    parallelix-node // start
    $ parallelix-node start --gpu
    > node 7 · ACTIVE · heartbeat 12s · max-jobs 8
    > 12:01:03 dispatch inf-0c9a · ai-inference
    > 12:01:09 execute cuda + onnx · t=5.1s
  5. // 05

    ACCRUE

    There is no per-task payout. Once the PoE commitment is validated, the completed work counts toward your uptime. The contract computes stake x tier x uptime and streams rewards per-second while your node stays online. Accrued $PRLX is claimable anytime via the Console.

    parallelix-node // accrue
    > 12:01:09 proof inf-0c9a · sha256 e7b2...
    > 12:01:14 validate redundant-sample ok
    > 12:01:14 uptime credited · rewards streaming
HARDWARE TIERS

MINIMUM TO RECOMMENDED.

Hardware tier determines your reward multiplier and the workload classes you qualify for. The network routes AI inference to qualified nodes. Tier declarations are verified at attach time by the coordinator.

reward multipliert1=1.0xt2=1.5xt3=2.5xt4=3.0x·applied as stake × tier × uptime in the per-second reward stream
  • // 01

    EDGE

    gamer · dev

    t1 · 1.0x
    cpu
    8c · avx2
    gpu
    rtx 3060 · 12 gb
    ram
    32 gb · ddr5
    net
    100 mbps
    storage
    100 gb ssd
    os
    linux
    lanes
    ai-inferencerender-2ddata-augoracle
  • // 02

    WORKSTATION

    studio · ml engineer

    t2/t3 · 1.5x / 2.5x
    cpu
    16c · avx2
    gpu
    rtx 4090 · 24 gb
    ram
    64 gb · ddr5
    net
    1 gbps · static
    storage
    500 gb nvme
    os
    linux / macos / windows
    lanes
    ai-inferencesimulationrender-3dzk-prove
  • // 03

    DATACENTER

    operator · datacenter

    t4 · 3.0x
    cpu
    64c+ · avx2
    gpu
    h100 / a100 · 80 gb
    ram
    256 gb+ · ddr5
    net
    10 gbps · static
    storage
    2 tb nvme
    os
    linux
    lanes
    ai-inferenceai-trainingsimulation-hpczk-heavy
Continue

Read the manual. Run a node.

The full operator spec lives in the docs. Install the binary, run parallelix-node setup, stake in the Console via the prefilled link, then start the daemon. Uptime earns. No per-task billing.

build
v2.0.0
runtime
node + ollama
signed
secp256k1
install
live
Read the operator manual$PRLX mechanics