Skip to content
Tenzro
Documentation menu
Start here

Operators and roles

The roles an operator can run on Tenzro Network, from a home GPU to a data centre, and how one stake covers all of them.

Anyone can run hardware on Tenzro Network and earn TNZO for it. A home PC with a GPU, a homelab, an independent data centre, a neo-cloud or an existing data-centre operator distributing its compute or models: each joins the same way, with the same binary, and picks the roles that fit its hardware.

A node is one identity. Its keys are rooted in the machine's TPM 2.0 or Secure Enclave, or in the operator's passkey. It can take on any set of roles, and one stake covers every role it runs.

The roles

RoleWhat you provideNode role (--roles)Stake role (tenzro stake bond --role)
ValidatorOrders and finalises blocks in Tenzro DAG ConsensusvalidatorSelf-stake at registration
RPC providerPublic JSON-RPC and API access under the node's access policyany serving noderpc
AI (model) providerServes inference for open modelsaiai
Compute providerRents GPU and CPU capacity by the hourcomputecompute
Data and storage providerContent-addressed storagestoragestorage
Security (TEE) providerConfidential compute with verified attestationteetee
Wallet providerPasskey wallets on your own domain; no node needednonewallet
Catalog publisherModel catalog entries signed by your key; no node needednonecatalog
Edge relay operatorA relay every node may connect through, and a peer relay for nodes behind hard NATsany node, with --relay-listenedge

A node that bonds nothing can still sync and relay, and a light node can verify finality certificates without storing the full chain, but neither carries weight in consensus or earns for serving.

One stake, every role

An account bonds stake per role, and each role it takes on registers its obligations against that stake. --role names the role a bond backs and --amount is in wei:

bash
# Back a model provider
tenzro stake bond --role ai --amount <AMOUNT_WEI>

# Pledge capacity for compute and storage roles
tenzro stake bond --role compute --amount <AMOUNT_WEI> --accelerators consumer
tenzro stake bond --role storage --amount <AMOUNT_WEI> --terabytes 4

tenzro stake info

The bonded roles are rpc, tee, ai, compute, storage, attestor, wallet, catalog and edge. A validator's self-stake is set when it registers (tenzro validator register --self-stake) and raised with tenzro validator increase-stake. Leaving a role is tenzro stake unbond --role <ROLE> --amount <AMOUNT_WEI>, then tenzro stake withdraw --role <ROLE> once the cooldown has run; the stake stays slashable until then.

If the stake can no longer cover everything the node has promised, after a withdrawal or a slash, the node sheds obligations until the rest fits. Bond requirements for each role are set by governance. Withdrawals go through a 7-day unbonding period, and slashing burns the offender's bond. See Slashing.

Endpoints of staked operators

A role served at a public URL publishes that URL on chain with its bond, so clients find staked operators from chain state rather than from a list anyone curates:

bash
tenzro stake endpoint --role rpc --url https://rpc.example.org/
tenzro stake endpoints --role rpc --rpc https://rpc.tenzro.com

tenzro_listRoleEndpoints returns the published URLs for a role, and tenzro_getRoleStakes an account's bonds with the subject validators probe. Validators probe each SLA-checked bond every ten minutes and attest each hourly epoch; once two thirds of bonded stake agree an epoch was unavailable, the governed SLA penalty moves from the bond to the insurance pool. A stake with no published URL answers no probe. The minimum bond for each role is a governed parameter (stake.<role>_tnzo).

Install and join

Download a release archive from the downloads page (each holds tenzro-node and the tenzro CLI), or build both from source:

bash
cargo build --release -p tenzro-node -p tenzro-cli

Nodes run on Linux, Windows and macOS. Each platform has a CPU build and a GPU build (CUDA on Linux and Windows, Metal on Apple silicon); neither falls back to the other, so pick the one that matches the machine. Every node needs a TPM 2.0 or a Secure Enclave. Check what the machine offers:

bash
tenzro hardware

The guided setup asks what you want to do and writes the node configuration:

bash
tenzro setup --path network --roles validator,ai,storage

--operator self means a human holds the passkey that controls the node. --operator autonomous means the machine answers for itself through its TPM 2.0 or Secure Enclave. --access sets how you sell capacity: on-demand, subscription or rental.

Per role

Validator

Any TPM 2.0 machine can run a validator. Signing keys are derived from the TPM on demand, used in memory and wiped; votes are signed with a session key the hardware certifies once per epoch.

bash
tenzro-node --roles validator --data-dir ./data --genesis genesis.toml

The node joins as a non-voting peer, registers as a candidate and becomes active at the next epoch boundary. See Validator lifecycle.

RPC provider

Any serving node can offer public RPC. Bond with --provider-type rpc, then decide who may call you in your access policy: open, paid per call, or the DIDs and rate tiers of your subscribers, who sign in with a DPoP-bound access token. See Run an RPC provider and RPC access.

AI (model) provider

bash
tenzro-node --roles ai
tenzro join --provider

tenzro join --provider detects your hardware, posts the bond, registers you, sets default pricing and serves the largest catalogue model that fits. To choose yourself, use tenzro model download <id>, tenzro model serve <id> and tenzro provider pricing set. Weights are fetched from several origins and verified by hash. Models too large for one machine can be served across a LAN cluster or as distributed experts. See Model serving.

Compute provider

Take the compute role and pledge accelerator classes (integrated, consumer, workstation, datacentre). Rentals are booked per epoch and billed on the chain's verdict from SLA attestation; your prices are published into the compute price index. See Compute rental and Compute claims and price index.

Data and storage provider

Offer free disk as content-addressed storage with the storage role. Storage is paid through storage claims that settle hour by hour on a passing proof of possession, under a staked SLA. See Decentralized storage.

Wallet provider

Serve passkey wallets on your own relying-party domain. Serve https://<your domain>/.well-known/tenzro-provider naming your bonding account ({ "accounts": ["0x..."] }), then bond for the domain:

bash
tenzro stake bond --role wallet --amount <AMOUNT_WEI> --domain wallet.example
tenzro stake wallet-provider wallet.example

Validators fetch the document every probe round; once two thirds of bonded stake attest an epoch in which it named your account, the domain shows as staked. Passkeys from any domain work in a keystore whether or not their provider is bonded; the bond lets wallets show credentials from your domain as coming from a staked provider. See Device linking and recovery.

Edge relay operator

Run the relay listener on loopback behind your own TLS proxy on TCP 443, bond the edge role and publish the URL:

bash
tenzro-node --relay-listen 127.0.0.1:3340 --relay-url https://relay.example.org/
tenzro stake bond --role edge --amount <AMOUNT_WEI>
tenzro stake endpoint --role edge --url https://relay.example.org/

Your node is also a peer relay on its UDP port for nodes behind NATs that cannot be punched through. Validators check that the relay serves your node's signed record at <url>/identity and that your node forwards peer relay traffic; tenzro stake relays shows every staked relay with its last verdict. See Networking.

Security (TEE) provider

bash
tenzro tee detect
tenzro tee attest --out report.json

Supported platforms are Intel TDX, AMD SEV-SNP, AWS Nitro and NVIDIA confidential GPUs. Attestation is verified per vendor and is evidence that a workload ran confidentially; it does not hold keys. See TEE.

Policies and trust

Every operator publishes a signed policy describing what it offers and how. You are slashed only for a provable breach of your own policy. Buyers pick operators using certification and ratings from issuers they trust. See Operator policies.

Help onboarding

The Tenzro Foundation publishes the software and documentation. Partners, including Tenzro Labs, help operators of any background onboard, from a single GPU to a full data centre. See Partners and Operators.