ogpu 0.2.3

OpenGPU SDK for distributed AI task deployment

OpenGPU Python SDK

Python SDK for the OpenGPU Network distributed AI compute marketplace.

pip install ogpu

Overview

The SDK has two sides:

• ogpu.chain — ChainConfig, ChainId, nonce utilities, ABI loader (re-exported at top level: from ogpu import ChainConfig)
• ogpu.client / ogpu.protocol — interact with the blockchain: publish sources and tasks, manage vault staking, monitor events
• ogpu.service — framework for source developers to expose task handlers inside Docker containers

Quick Start

Publish a Task

import time
from web3 import Web3
from ogpu import ChainConfig, ChainId
from ogpu.client import TaskInfo, TaskInput, publish_task

ChainConfig.set_chain(ChainId.OGPU_TESTNET)

task = publish_task(TaskInfo(
    source="0xYOUR_SOURCE_ADDRESS",
    config=TaskInput(function_name="predict", data={"prompt": "hello"}),
    expiryTime=int(time.time()) + 3600,
    payment=Web3.to_wei(0.01, "ether"),
))

print(task.address)             # contract address
print(task.get_status())        # TaskStatus.NEW
print(task.get_source().address)  # navigate to the source

Instance Classes

Every on-chain entity has a live proxy class. Methods hit the chain on every call — no stale cache.

from ogpu.protocol import Source, Task, Response, Provider, Master

task = Task.load("0xTASK_ADDRESS")   # validates existence on-chain
task.get_status()                     # TaskStatus
task.get_attempters()                 # list[str]
task.get_confirmed_response()         # Response | None (chain-only)
task.cancel(signer=my_key)            # Receipt

snap = task.snapshot()                # frozen capture of all fields
print(snap.status, snap.payment, snap.attempt_count)

Vault Operations

from ogpu.protocol import vault

vault.deposit("0xBENEFICIARY", amount=10**18, signer=my_key)
vault.lock(amount=5 * 10**17, signer=my_key)
vault.get_balance_of("0xADDRESS")    # int (wei)
vault.is_eligible("0xADDRESS")       # bool

Event Watching (Async)

import asyncio
from ogpu.events import watch_attempted

async def monitor(task_addr: str):
    async for event in watch_attempted(task_addr):
        print(f"Attempt from {event.provider} @ block {event.block_number}")

asyncio.run(monitor("0xTASK_ADDRESS"))

Custom RPC

from ogpu import ChainConfig

ChainConfig.set_rpc("https://my-private-node.example")
ChainConfig.get_rpc()     # current URL
ChainConfig.reset_rpc()   # restore default

Architecture

ogpu/
  chain/        # ChainConfig, ChainId, Web3Manager, nonce, ABI files
  types/        # enums, errors, Receipt, metadata
  protocol/     # low-level 1:1 contract wrappers
    nexus, controller, terminal, vault
    Source, Task, Response, Provider, Master
  client/       # high-level client workflows
  events/       # async event watchers (the one async island)
  service/      # source developer framework (separate concern)

Protocol layer — every Solidity view has a Python accessor, every state-changing function has a writer. Signer management via per-call signer= parameter with role-based env-var fallback (CLIENT_PRIVATE_KEY, PROVIDER_PRIVATE_KEY, MASTER_PRIVATE_KEY).

Instance classes — Source(addr), Task(addr), Response(addr), Provider(addr), Master(addr). Stateless — only self.address is stored. Task.load(addr) validates existence eagerly.

Events — async generators in ogpu.events. HTTP filter polling, not WebSocket. Rest of SDK is sync.

Requirements

• Python >= 3.10
• web3 >= 7.0

Documentation

Full docs: https://opengpu-network.github.io/sdk-ogpu-py/

Writing a Task Handler (ogpu.service)

import ogpu.service
from pydantic import BaseModel

class MultiplyInput(BaseModel):
    a: int
    b: int

class MultiplyOutput(BaseModel):
    result: int

@ogpu.service.expose()
def multiply(data: MultiplyInput) -> MultiplyOutput:
    return MultiplyOutput(result=data.a * data.b)

ogpu.service.start()

License

Released under the MIT License — Copyright © 2026 OpenGPU Network.

You're free to use, modify, distribute, and embed ogpu-py in commercial or open-source projects. The only requirement is that the copyright notice and license text travel with the source.