Skip to content

Installation

Requirements

  • Python 3.10+
  • Access to an OpenGPU RPC endpoint (the SDK ships with defaults for mainnet and testnet)
  • A funded wallet on the chain you target (testnet tokens for testing)

Install

pip install ogpu

pip install "ogpu[examples]"

Adds Jupyter so the scenario notebooks under examples/scenarios/ can be opened and re-run locally.

git clone https://github.com/OpenGPU-Network/sdk-ogpu-py.git
cd sdk-ogpu-py
pip install -e ".[dev]"

Verify

import ogpu
print(ogpu.ChainConfig.get_current_chain().name)  # OGPU_MAINNET
print(ogpu.ChainConfig.get_rpc())                 # https://mainnet-rpc.ogpuscan.io

If the import succeeds and the RPC URL prints, you're ready.

Environment variables

The SDK reads private keys from env vars, looked up by role. None are required at import time — you only need the one matching the role you act as.

Variable Used by Description
CLIENT_PRIVATE_KEY ogpu.client Publish sources, publish tasks, confirm responses
PROVIDER_PRIVATE_KEY provider-side writes announce_master, register, attempt
MASTER_PRIVATE_KEY master-side writes announce_provider, remove_provider, set_agent
AGENT_PRIVATE_KEY ogpu.agent Scheduler role delegated by a master

The SDK searches for a .env file in this order on import:

  1. Current working directory (./.env)
  2. User home (~/.env)
  3. SDK install directory (fallback)

Any values already set in the environment take precedence.

Vault operations need explicit signers

Vault writes (deposit, withdraw, lock, unbond, claim, cancel_unbonding) require signer= as a keyword argument. There is no env-var fallback — this is intentional, to prevent accidental deposit/withdraw from the wrong account.

Pick a chain

Default chain is mainnet. Switch to testnet explicitly:

from ogpu import ChainConfig, ChainId

ChainConfig.set_chain(ChainId.OGPU_TESTNET)

For a custom RPC endpoint (private node, local fork), see custom RPC.

Next

  • Quickstart — publish and confirm a task in a handful of lines
  • Concepts — how the packages and layers fit together