Skip to content

README.md

Latest commit

 

History

History
104 lines (70 loc) · 7.11 KB

README.md

File metadata and controls

104 lines (70 loc) · 7.11 KB
Raw

InfiniteZero Network — DevNet

Important

You're on develop, the development branch. It holds the next version of the protocol (DevNet 2.0, not deployed yet) and is where all contributions go.

Want to use the live DevNet? DevNet 1.0 on Optimism Sepolia runs the code on main. Start with its README and Getting Started guide, and install dincli 0.2.0.

CI License: Apache-2.0 Good first issues Telegram Built on Ethereum Founded at Oxford

The internet gave everyone a voice.
Open source gave everyone a tool.
Yet AI learns from only a fraction of the world's data.

InfiniteZero lets AI learn from the rest — where it lives, for everyone.

What is this?

InfiniteZero is an open protocol for training shared AI models on data where it lives. Participants train models on their own devices with federated learning and share model updates, not raw data. Smart contracts on Ethereum coordinate each training round, check the work, and reward the people who do it.

This repository is the DevNet: the smart contracts, the dincli command-line tool, the reference model services, and the documentation for everyone building on or running the network.

The protocol was previously called DIN (Decentralized Intelligence Network). You'll still see that name in the code, for example in dincli, until the rename to InfiniteZero (#281) is complete.

Where we are today

  • DevNet 1.0 is live and experimental, on the Optimism Sepolia test network, using test ETH. It runs the code on main. Validator nodes run 24/7 from Japan to Canada. Expect bugs, changes and resets.
  • DevNet 2.0 is next. This develop branch holds it: commit-reveal scoring, the reward engine, fee routing and emissions, on upgradeable contracts. It isn't deployed yet.
  • One tested example so far: an MNIST image model runs end to end. Other model types are supported by design but not yet tested.
  • Next: a public testnet, then mainnet with a fully decentralised fair launch, with no ICO and no VC allocation.
  • Privacy: raw data stays on devices. Optional differential-privacy settings exist but are off by default and still being strengthened, so we make no formal privacy guarantee yet.
  • Governance: decisions are made in the open today; on-chain community governance is planned after mainnet.

Start here

You want to… Go to
Run a node or join a training round on the live DevNet 1.0 Getting Started guide on main and dincli 0.2.0
Try the DevNet 2.0 code on develop Getting started · Setup guide
Train your own model on the network What can I train?
Evaluate InfiniteZero for your community or organisation Partner introduction
Contribute code, docs, research or translations Good first issues · Contributing guide · Roadmap

All documentation: Documentation/

What's in this repository

Folder Contents
foundry/ Solidity smart contracts (staking, slashing, rewards, emission, fair-launch distribution) and their test suite
dincli/ The dincli command-line tool used by validators, clients and model owners
cache_model_0/ The reference MNIST model and its training, auditing and aggregation services
Documentation/ What exists today: public/ for participants, technical/ for code readers and auditors
Developer/ Plans, designs, tasks and process docs, including the roadmap and backlog

Since October 2026, every change to develop needs an approving review from someone other than its author (MAINTAINING.md), and CI runs the Solidity test suite and the Python unit tests on every push.

Contributing

The network is open and experimental. We work on the develop branch, and no contribution is too small: documentation, testing, tooling, smart contracts, privacy research and translations all matter.

  1. Pick a good first issue or a help wanted issue, and comment to claim it.
  2. Read the contributing guide and development setup.
  3. Branch off develop and open your pull request against develop.

Questions? Ask in Telegram or say hello in Discussions.

Found a security issue? Please don't open a public issue. See our security policy.

Learn more and support the work

If you use this software in research, please cite it using the "Cite this repository" button on the right.

InfiniteZero Foundation — open AI infrastructure, built by everyone, for everyone.
Open protocol · Built in the open · Apache-2.0 · © 2026