The REI Framework: Connecting Artificial Intelligence and Blockchains
The Rei framework is a universal translator born out of the need to facilitate how AI and blockchains can communicate.
One of the main issues with creating AI agents is ensuring they are flexible enough to learn, iterate, and grow while being consistent in the output produced.
Rei creates a framework to share structured data between AI and blockchains, allowing AI agents to learn, improve, and retain a knowledge base of experience and understanding.
This enables the development of AI systems able to:
Understand context and patterns and generate insights
Transforming insights into executable actions while benefiting from the transparency and reliability of blockchains.
The Issue at Hand
There is a fundamental division between the properties of AI and blockchains:
Blockchain computation is deterministic: “Every operation, every calculation, and every state transition must produce identical results across all nodes in the network” to ensure:
Consensus: New blocks are verified by each node that agrees on the content
State Verification: The blockchain state should always be verifiable from inception. New nodes joining the network should arrive at the same state as all other nodes.
Smart Contract execution: all nodes should produce the same outputs given the smart contract inputs
AI Systems generate “probability distribution across all possible outcomes”.
Their results are probabilistic and might produce different results every time they run (due to sampling techniques, different numerical optimization procedures, arithmetic variations, or hardware-specific implementations)
Context Dependency: AI systems rely on the context provided (e.g., training data history, model parameters, temporal and environmental conditions)
Resource Intensity: computations require intense computational power, complex matrix operations, extensive memory, and specialized hardware.
This gives rise to many conflicts with to regard compatibility:
Probabilistic vs Deterministic Data:
How do we convert probabilistic outputs to deterministic results?
When and where should this conversion occur?
How do we preserve the value of probabilistic insights in a deterministic environment?
Gas costs: the computational requirement of AI networks would make gas costs prohibitive.
Memory limitations: Blockchain environments have strict memory limitations, which would not be able to sustain the requirements of AI models.
Execution Time: block times limit execution, affecting the speed of AI models.
Data Structures Integration: AI models use complex data structures which are hard to integrate within blockchains’ storage patterns
Oracle Problems (Verification Requirements): how can we verify the validity of AI computations? AI systems require rich context and low latency.
How do we connect AI agents and blockchains?
Rei’s approach proposes taking the best of both worlds.
Instead of forcing these two systems together, Rai focuses on becoming a universal translator, with a translation layer allowing AI and blockchains to communicate and interact.
Need to:
Let agents think and learn independently
Record AI agent insights into precise and verifiable blockchain actions.
The first implementation of this framework is Unit00x0 (Rei_00 - $REI), currently trained to become a quant.
The cognitive architecture powering Rei takes the form of a four-layered process:
Thinking Layer: the agent processes and gathers raw data (e.g., chart data, transaction history, user behavior), looking for patterns.
Reasoning Layer: uses patterns and adds context to it. While the thinking layer is used to spot patterns, the reasoning layer connects them with other information (e.g,. day and time, historical trends, and market conditions).
Decision Layer: decides what actions to take based on the contextualized information the Reasoning Layer provides.
Action Layer: decisions translate into deterministic blockchain actions.
At the foundation of the Rei framework, there are three distinct pillars:
An Oracle (acting as a neural pathway): translating different AI outputs into consistent results stored onchain.
ERC Data Standard: expanding blockchain storage to store complex patterns and allowing on-chain data to maintain the rich context provided by the Thinking and Reasoning Layer (from probabilistic data to deterministic execution).
Memory System: allows Rei to learn over time and access previous outputs and experiences.
Here’s what these interactions look like:
The Oracle Bridge identifies patterns
ERCData stores patterns
Memory systems maintain the context to understand these patterns
Smart contracts can access and act on this accumulated knowledge
Thanks to this architecture Rei agent is already able to carry out in-depth analysis on tokens combining:
Rei is not only able to analyze the data provided but also develops a deeper understanding. This is achieved by storing information about agent experiences and insights directly onchain. This way, they become part of her knowledge base and can be retrieved and leveraged to refine her decision-making framework and experience further.
Sources and Outputs of Rei data: Plotly and Matplotlib library (for charting), Coingecko, Defillama, onchain data, and Twitter for social sentiment.
With the Quant V2 update, Rei can run several analysis formats:
Project Analysis: now, with the addition of quantitative metrics and sentiment data. Candlestick chart + Engagement Chart + Holder Distribution (& PnL).
https://x.com/unit00x0/status/1874854497249361942Inflows and Outflows Analysis: monitoring price and volume for onchain trending tokens, comparing it against the inflows and outflows. https://x.com/unit00x0/status/1874833132211290367
Engagement Analysis: assessing projects’ general engagement (instant vs 24 hours before) and the relative price change, showing the correlation between recent information and engagement performance. https://x.com/unit00x0/status/1874594348706173248
Top Category Analysis: bottom volume and top transactions for single categories, highlighting how projects perform against their category.
The first chart shows the volume on the bottom and the number of transactions on the top, going in deep later on single categories, highlighting changes in metrics for a single project against its category.
https://x.com/unit00x0/status/1874848100092649742
As of January 2025, Rei can also buy and sell tokens on-chain, as she’s provided with a smart contract wallet using ERC-4337.
The smart contract delegates the operations to Rei with a user signature; this way, she can manage her portfolio autonomously.
This is Rei's EOA wallet (signature wallet) https://basescan.org/address/0x3BC4c3A2a2Fa5ad20a2B95B18CA418D06A360cB This is Rei's Smart Wallet (account abstracted wallet) https://basescan.org/address/0xf6835acc8d2b51e5d47632ca8954bfee9a0ce49c
Use Cases
The Rei framework goes beyond financial applications for AI agents, opening up their use across a wide range of use cases based on:
User-Agent Interactions: Content Creation
Market analysis: Supply Chain, Logistic
Creation of Adaptive Systems: Governance
Risk assessment using Contextual analysis: Healthcare
What’s next for Rei?
Token-gated alpha terminal for custom analyses
A platform for people to develop their agent
More on Rei:
https://x.com/0xlukasinho/status/1876677684547535158
https://x.com/0xreisearch/status/1876752172492451888
https://x.com/0xreisearch/status/1875007578398212174
https://x.com/0xreisearch/status/1874866413191729604