The Story of Rakis
Rakis is a project born out of our fascination with the intersection of artificial intelligence and decentralized systems. The idea of creating a permissionless, decentralized network for running AI inference had been brewing in our heads for quite some time. We belive the unlocks it offers to builders and users alike are step function improvements over current status quo.
The journey of Rakis began with a simple question: "What if we could run large language models in a decentralized and trustless manner?" This question quickly evolved into a deeper exploration of the challenges and possibilities of combining AI and blockchain technology.
While the idea of decentralized AI inference is not entirely new, most existing solutions technologies that don't exist yet (without working towards building them), or skip over the consensus question entirely. "How are you planning to have trusted verifiable inference in a decentralized system?" is a good question to ask.
The vision for Rakis was to create a truly decentralized and permissionless network, where inference could be performed without relying on any centralized authority or trusted parties. This meant addressing several key challenges, such as:
- Ensuring the integrity and determinism of AI inference results in a decentralized environment.
- Enabling seamless integration with existing blockchain networks and smart contracts.
- Providing a scalable and efficient infrastructure for running large language models and other AI models.
The name "Rakis" is an omage to the new name for Arrakis under Leto. Nodes in the Rakis network are called Synthients, after the race of embodied AIs in the Matrix universe.
Step 1: Establish a Robust P2P Network
The first step in building Rakis was to establish a robust peer-to-peer network that could facilitate the exchange of inference requests, results, and other data among nodes. This involved integrating with multiple P2P networks, such as Nostr, GunDB, and NKN, to ensure redundancy and resilience. In this experimental stability test phase, we want to stress test each p2p network to find the most optimal solution.
Step 2: Develop a Novel Consensus Mechanism
At the core of Rakis lies a novel consensus mechanism designed specifically for decentralized AI inference. This mechanism leverages embedding-based consensus and a commit-reveal system to ensure the integrity and determinism of inference results, even in the presence of Byzantine actors.
Step 3: Enable Browser-Based Inference
To truly democratize access to AI inference, Rakis was designed to run entirely in the browser, without relying on centralized servers or infrastructure. This involved integrating with cutting-edge technologies like WebGPU and WebAssembly to enable efficient inference on client devices.
Step 4: Integrate with Blockchain Networks
To facilitate seamless integration with existing blockchain networks and smart contracts, Rakis incorporates support for connecting AI inference capabilities to various chains, including Ethereum, Arbitrum, and Solana (planned). This opens up a world of possibilities for decentralized applications (dApps) and autonomous agents.
Read more about the creation and motivation behind Rakis.