The AI Dispatch #27 - Same Old, $REI Network, Hyperliquid AI?

The AI Dispatch #27 - Same Old, $REI Network, Hyperliquid AI? — research from Castle Labs.

Castle Labs28 July 20253 min read

Welcome to week 27 of our AI dispatch!

It's great times all over, and we're experiencing this week with waves of enthusiasm. After a slight decline over the past few weeks, Bitcoin and the broader crypto market appear to have regained strong momentum this week, driven by more tariff deals signed at a macroeconomic level.

This week, we have the pleasure of hosting a guest piece on $REI!

Let's start with our usual market overview.

Market Overview

One of the hottest topics of the week has been @TheoriqAI raise through @KaitoAI and the @Almanak__ raise through @legiondotcc.

This highlights both how niche AI projects are showing demand beyond their niche, as well as a broader development hinting at a greater focus on launchpads and community raises.

At the same time, the AI market as a whole has still been lagging behind other, more established tokens, which have experienced strong growth in the past couple of weeks.

Among those, we absolutely have to mention what REKT has done, since it has been literally skyrocketing on the charts, but even more, it has shown that an organic model of community building is possible for NFTs, as well as expanding the core business beyond that through AI and real-life activations with the REKT drinks.

On Rei Network: AI Beyond LLMs - Multimodal Synthetic Brain Architecture

@ReiNetwork0x is an AI lab focused on advancing artificial intelligence through science-driven innovation.

Their main AI architecture, “Core”, is a multimodal synthetic-brain architecture that solves the three fundamental problems plaguing today's LLMs:

Stateless memory
Frozen post-training knowledge
Hallucinations.

Article figure

To enable user interaction with Core, Rei has a working product that allows users to create and interact with instances of Core, known as "Units".

Where are they now?

The team has shipped three major Core iterations and numerous product updates in the past 7 months. The most recent Core iterations unlocked training at inference time, enabling agents to learn and adapt from usage and behaviour memory.

Unlike LLMs, which would require costly offline fine-tuning, Core enables Units to learn explicitly and implicitly through user interactions.

Rei beta users have shared various usage on X, notably esports training from Team Nemesis, who qualified for the Dota 2 World Championship, and perps trading with the hybrid protocol by Ecliptic Fund.

Article figure

Their latest crypto data integrations (Defillama, Elfa, Birdeye, Nansen, Defilama) unlock various onchain analysis of wallets, protocols, and transactions.

Article figure

What’s Next?

Despite being in closed beta, the product has experienced exponential growth in usage, with beta testers noting its ability to replace regular LLM usage.

Article figure

n addition to upcoming Core iterations, the team has mentioned scaling architecture, academic publications, open beta, and API distribution ecosystem. The team's aggressive release schedule delivers a steady stream of powerful upgrades, positioning Core not just for what it is today, but for what it can become in the coming months.

Curious to try? Apply for Beta Access:

https://reisearch.box/waitlist