TPan here! I’m still on the second leg of my honeymoon so I won’t be writing my usual pieces. However, I do have some great content for you from someone else in the space.

Today we have Chris Sotraidis and he’ll share some insights on the intersection of AI and web3. Chris and I hit it off in the early days of my web3 and writing journey, and we’ve become good friends, even going to conferences together. Chris has stayed on top of all things AI as it has dominated headlines over the past year, and am grateful for his willingness to share his insights and observations with us. 

Chris is a writer at Spatial Awareness and an active contributor in the web3 space. If you’d like to learn more about what he’s up to, you can reach out to him here.

This is your last chance. After this, there is no turning back. You take the blue pill - the story ends, you wake up in your bed and believe whatever you want to believe. You take the red pill - you stay in Wonderland and I show you how deep the rabbit hole goes.

Hey ya’ll, it's Chris. Back in December, I wrote an AI and Web3 in 2024 article, highlighting some of my thoughts and predictions about the space for the year. The first section I wrote highlighted agent-based intelligence being the core of web3’s AI evolution for this year, and that is proving to be largely correct (with DePIN blowing up as well). There’s a number of reasons for this, and the convergence of these technologies is both exciting and leading to creation of many interesting projects. This is leading to what is being called the Internet of Agents (IoA). 

The project I want to highlight and write about today is called Morpheus AI. The yellow paper, authored anonymously by Morpheus, Trinity, and Neo, spells out the technical details of the Morpheus full node, smart contract, and related proofs. Morpheus is marketed as the ‘first peer-to-peer network of personal AIs’ known as smart agents. You can think of smart agents as being an unlock for usage in web3, where regular users can talk with their agent in natural language and have the agent take action appropriately.

So, let’s get into it!

To start, let’s try to visualize web3 (or just the web) in the near future. Close your eyes and take a deep breath. As AI usage ramps up, so does the desire for solid content provenance. Blockchain, along with continued advancements in cryptography in general, form the backbone of the internet. Users on the internet want to know if who they’re talking to is really who they think they’re talking to, and they want to be assured that the content they consume online is really what they think it is.

Today, cryptographic protocols like TLS already provide seemingly invisible assurance when corresponding via email and VoIP, but scamming using AI (like deepfakes and voice replication) is on the rise. As web3 (and the internet in general) becomes increasingly complex, helpful, auditable, and personal AI companions will be used to make navigating the web both safer and easier.

I know it seems hard to believe, but there was a time before browsers for navigating the internet. CLI, BBS, and FTP were largely used to transfer files and type commands directly into a terminal. They were usable, but they weren’t user-friendly, and the barrier to entry and actually using the internet was vastly less common. The creation of the modern world wide web and the introduction of the web browser made the Internet what it is today, even if the first browsers look primitive by today’s standards.

It is possible we are living in a similar bubble with modern UI. You have to click, navigate through (hopefully well-designed) menus, and find your intention. This isn’t how humans naturally think. We just want to do things. I want to take a car ride from Orlando’s airport to my house. I want to order a large pineapple pizza. I want to look at the cat photo you just sent me. A change even bigger than the upgrade from CLI to browser is on the horizon, and it will make a lot of things a lot easier to do. I’m talking about the intersection of LAMs, personalized AI, IoA, and a fundamental shift in human-computer interaction (HCI).

At the beginning of the year, Rabbit blew up, marketing itself as a device with a personalized operating system through a natural language interface. The jury is still out on if Rabbit can deliver on its promise of radically transforming our lives, but the form factor (teenage engineering), and the UI flow excited people tremendously. Creating a large action/agentic model (LAM) that can infer human actions by navigating a traditional UI and perform them quickly, as if a human is at the wheel, is groundbreaking. Accessibility aside, A LAM system could make tedious or otherwise frustrating UI experiences turn into a simple voice command. Rabbit is a physical hardware example of this idea (I’ll be reviewing it in a couple weeks once it arrives in the mail), but this notion extends to the world of web3, a place which is also at times tedious, inaccessible, or difficult to understand. You can think of Morpheus as being a decentralized peer-to-peer network of general-purpose AIs. It doesn’t take much imagination to envision how a tool like this could at first be useful for reviewing smart contracts, but its breadth could be extended to look something like the idealized version of MultiOn (another agent-based API).

What does that look like in practice? Imagine the following:

In 2026, TPan decides to order his favorite cheese puffs from Amazon. He summons his trusty Web3 agent, "SnackBot," which is powered by a large action model (LAM). SnackBot, having learned TPan's preferences, quantity, and budget from previous interactions, immediately begins searching for the best deal on cheese puffs. Using its advanced AI algorithms, SnackBot swiftly identifies the optimal choice and initiates the purchase using TPan's linked cryptocurrency wallet. The entire transaction is recorded on a blockchain, ensuring transparency and accountability.

Within seconds, the order is complete, and TPan receives a notification confirming the purchase. He smiles, knowing that his cheese puffs are on their way, thanks to the convenience and efficiency of his Web3 agent, SnackBot.

Wow, talk about convenience! TPan didn’t even have to tell the bot his income, quantity, or preferences, the model simply remembers and makes the choice that TPan would have made. More time left over to day-trade on Solana! This is simplifying things, so I encourage you to read Rabbit’s research page to get a better understanding of how it works in comparison to an LLM.

You can think of Morpheus AI, or at least the idealized version of it, as being a web-oriented, on-chain LAM —  smart agents with long-term memory. Extend the analogy above of TPan ordering cheese puffs on Amazon to you performing a more complicated action within the wild world of web3.

But, before we dig into Morpheus, it is important to understand the philosophy and theories that have driven the creation of this project. They are largely inspired by Vitalik Buterin’s d/acc (decentralized accelerationism) and Marc Andreesen’s Techno-Capitalist Machine.

Don’t take my word for the TL;DR here. Reading the source material is essential! The greatest new decentralized projects, especially anon ones, are (often) highly influenced by Vitalik and Marc (and also Beff Jezos). Below I’ve briefly outlined d/acc and the Techno-Capitalist Machine, two ideas that have clearly influenced Morpheus.

Vitalik Buterin's d/acc (decentralized accelerationism):

D/acc represents a balanced approach to technological progress, emphasizing decentralization, democracy, and defense. It seeks to harness the benefits of technological advancements while urgently calling for democratic and decentralized solutions to mitigate potential risks. D/acc focuses on technologies that contribute to a safer, more open, and freedom-respecting governance form. Plus, we need a counter to e/acc t-shirts and hats cropping up all over the Bay area.

Key aspects of d/acc:

• Prioritizes defensive technologies over offensive ones

• Advocates for robust physical infrastructure and open-source biomedical research

• Emphasizes enhanced computer security through cryptography and blockchain

• Develops tools to combat misinformation and enable effective decentralized governance

Marc Andreesen's Techno-Capitalist Machine:

The Techno-Capitalist Machine (TCM), as described in Andreesen's "Techno-Optimist Manifesto," combines technology and markets to create a self-perpetuating engine of material creation, growth, and abundance. This "techno-capital upward spiral" has been running for centuries, leading to increased specialization, trade, falling prices, and the highest levels of material living standards in history. The Morpheus project utilizes TCM as a framework for actually launching the open source project and deploying a native token.

Key aspects of the Techno-Capitalist Machine:

• Believes in the power of natural selection in the realm of ideas, with the best and most productive ideas winning out

• Advocates for accelerationism—the deliberate propulsion of technological development

• Seeks to ensure the fulfillment of the Law of Accelerating Returns, where technological advances feed on themselves, increasing the rate of further advance

• Focuses on maximizing innovation and capitalist forces to drive rapid technological advancement and material abundance

The inspiration isn’t mere postulation here, these writings are directly referenced in the README files within Morpheus’ Github. But, how do some of these lofty ideas actually translate into being actively implemented into the project? This is written about at length here.

So far, we’ve dug into the background info, emerging trends in AI (like LAMs vs LLMs), and influential ideas circulating on AI Twitter. Now, let’s dig into Morpheus. 

We can think of Morpheus as being a sophisticated network that enables decentralized agents, agents that are built, registered, and audited on Morpheus. The community builds the front-end interfaces (if necessary), with Morpheus connecting to web3 wallets, utilizing Filecoin for decentralized file storage. Agent stores (like Dapps) are created, providing access to audited agents so you don’t have to build your own. The best part: users can talk to their agents in natural language. The result is shared ownership and governance, building an autonomous agent economy. A huge shoutout to Davide Crapis for his very comprehensive article, The Internet of Agents – this article is a must-read if you find the notion of an internet filled with AI compelling, terrifying, or interesting.

That said, this is all happening at a time when big tech reigns supreme, and the best current choices for LLMs are solutions provided by OpenAI (essentially Microsoft), Google, and other solutions like Anthropic (now supported by Amazon). Cloud-based LLMs running on enterprise infrastructure offer faster response times, but they often monetize data, have censored models, and there’s limitations on what you can do. Easy to install and use open-source LLMs have been on the rise for a while, and they’re finally getting to a point where they nearly rival closed-source solutions (check out Llama 3 and run it locally to see what I mean).

Beyond avoiding censorship and handing over all your data to big tech, there are a slew of other things to be excited about when it comes to Morpheus’ tech stack.

Morpheus combines the following tech in a way that makes sense:

• Local Large Language Models (LLMs): Morpheus leverages locally hosted open-source LLMs like Llama and Falcon, which enable powerful natural language processing and generation capabilities without relying on remote servers, enhancing privacy and reducing latency.

• Web3 Wallet Integration: Morpheus integrates with the user's Web3 wallet for key management and to sign recommended transactions when interacting with the Smart Agent, enabling seamless interaction with decentralized applications, smart contracts, and cryptocurrencies.

• SmartContractRank Algorithm: Morpheus utilizes the SmartContractRank algorithm to score and recommend the best smart contracts to the user, helping them navigate the Web3 ecosystem effectively.

• Decentralized File Storage: Morpheus employs decentralized file storage systems, such as IPFS (InterPlanetary File System) and Filecoin, to store user data, prompts, and connected application information, ensuring data availability, resilience, and user control.

• Electron Framework: Morpheus leverages the Electron framework to package its software as a one-click install, making it accessible to non-technical users.

• Ethereum Smart Contracts: Morpheus utilizes Ethereum-based smart contracts, particularly on the Arbitrum L2 solution, to manage token rewards, staking, and other network functions, benefiting from Ethereum's security and the reduced gas costs of L2s.

• Zero-Knowledge Proofs (ZKPs): Morpheus plans to leverage zero-knowledge proofs for scalability and privacy, as Arbitrum and other L2 solutions continue to integrate ZK tech into their offerings.

• Fully Homomorphic Encryption (FHE): To protect user data when sending prompts to the Morpheus peer-to-peer network of compute providers, the software aims to leverage FHE versions of LLMs as they become available, ensuring privacy while maintaining computational efficiency.

These technologies work together in the Morpheus stack to create a decentralized, privacy-preserving, and user-friendly platform for powering agents. The simple install utilizing Electron makes this essentially a one-click install, with the ability to spin up new agents quickly and at zero cost. Developers are incentivized to contribute code to both Morpheus and the eventual agent app store, which hopes that this new protocol will become a go-to for agentic-savvy web3 users.

The tech is certainly interesting, and seeing all of this combined in one place is intriguing. But what does that look like in practice? What does installing and utilizing smart agents actually look like?

The project is new enough that I don’t have a demo to share. I would encourage those that are interested to check out the site and join the Discord. We can share diagrams and talk abstractly all day, but this project being relatively nascent means we can only make presumptions on how this will function in practice.

Here’s my take on what agents will look like in the near future:

• Autonomously monitor and analyze vast amounts of blockchain data, providing users with real-time insights, alerts, and recommendations based on their specified interests and/or holdings.

• Engage in complex, multi-step interactions with decentralized applications and smart contracts, automating tasks such as yield farming, liquidity provision, and token swaps based on predefined user preferences and risk tolerances.

• Collaborate with other Morpheus agents to pool resources, share insights, and execute coordinated strategies across multiple protocols and platforms, enabling users to participate in DAOs and collective investment vehicles.

• Continuously learn and adapt to user behavior and changing market conditions, leveraging ML to optimize strategies and maximize ROI.

• Serve as decentralized oracles, securely bridging off-chain data and real-world events with blockchain-based smart contracts.

• Provide seamless cross-chain interoperability, allowing users to manage and transact with assets across multiple blockchains through a single, unified interface powered by their Morpheus agent.

• Offer advanced privacy features, such as ZKP generation and fully homomorphic encryption, enabling users to participate in decentralized markets and interactions without revealing sensitive data or compromising their anonymity.

There it is. That’s what I believe you’d expect from functionality in the near-ish future if you choose to mess around with and dig into Morpheus more. Beyond that, tokenomics for the MOR token are top of mind.

How does the token work? What is the supply? I’ve tried to simplify and summarize this as much as possible below.

MOR tokens are rewarded daily to the key contributors of the Morpheus ecosystem:

• 24% to the Community (builders creating frontends, tools, and bringing users)

• 24% to Capital providers (funding for compute and code)

• 24% to Compute providers (equipment and power)

• 24% to Coders (intelligence behind the frontends, capital, and compute)

• 4% for protection funds

The MOR token supply is capped at 42 million, with a fair launch and no pre-mine or early token sale. The block reward starts at 14,400 MOR per day and gradually declines over 16 years until reaching zero. By then, fees from data, compute, capital, and coders will take over as the primary incentive.

Morpheus participants earn MOR tokens by providing proof of work (labor) and proof of stake (capital) to the network:

• Capital: Locking ETH in a yield-generating smart contract, with 50% of the yield used to buy MOR via an AMM and provide liquidity.

• Compute: Receiving MOR for user prompts and API calls served.

• Coders: Earning MOR for contributions merged into the Morpheus repo.

• Community: Receiving MOR for building frontends, tools, and engaging users.

To align incentives long-term, Morpheus features a tail emission of MOR tokens, calculated based on the number of burned tokens in the previous 5,833-day period. This emission is capped at 16% of the circulating supply, ensuring a sustainable and scarce token model.

Beyond all of this, I again encourage the reader to check out the README docs within Morpheus’ Github if you really want to deeply understand various aspects of how it will work, like the multisig setup, phased AMM deployment, testing plans, and Morpheus’ utilization of the Lumerin protocol to connect AI users with AI models.

Morpheus’ Discord is active, with the community actively discussing AI news, and keenly focused on agentic news as it relates to the space the community is building in. The token generation event is set for May 8th, with those wishing to participate in providing compute waiting until phase 2 launches in May/June. The community is working with Akash to create a template to make providing compute possible in a decentralized fashion. You can’t run a node on Windows, only Mac and Linux, as Ollama provides inference and is only available there. That may expand in the near future.

Morpheus is interesting to say the least. It is one of the projects that I kept saying I’d write about because it combines so many interesting ideas in a single place. Like Bitcoin, it was founded anonymously, and it is designed in a fashion that it will be usable or even active long after we’re all gone (unless you’re Bryan Johnson). The community is active and eager to answer questions. If this is intriguing to you, consider checking out Olas Network as well. An agent-based internet could be a blip on our path to true AGI, and along the way many terms and practices may be coined that will stick around once AGI is here to stay.

Personally, I’m excited for more agentic devices to hit the market soon, so I’ll plan on taking a closer look at them, reviewing them, and writing here. 2024 is the year of personal AI, and it is happening just as fast as we expected.

It’s a matter of when, not if for Web3 x AI to have its ChatGPT moment. Will Morpheus have a part in it? If not, it wouldn’t be surprising if it were something similar to it, but with NFTs instead of cheese puffs 😉

If you want to follow these and related topics closely, check out Spatial Awareness or reach out to Chris directly. 

See you soon!