What are AI agents

Artificial intelligence (AI) and Crypto (and ML, ZK, etc.) may seem like buzzwords for projects to raise funds at insane valuations. But, in reality, the synergy between these narratives is undeniable.

At the moment, a lot of on-chain transactions are already controlled by code (arbitrage, liquidation, and MEV bots). The next step is AI agents.

AI agents are computer programs that use artificial intelligence models to interact with apps and perform tasks on behalf of a user. These agents can be used for anything from personal assistants to financial managers.

AI agents can complete complex tasks for users and trade more effectively than humans due to their absence of bias, fatigue, and emotions. Additionally, their ability to recognize complex patterns and swiftly adapt strategies to changing market conditions surpasses human capabilities.

Blockchain technology provides the perfect environment for this technology to work and take new shapes with the help of the developing projects and their respective ecosystems.

Under the hood, AI agents make use of the following components. Refer to this list for all the buzzword definitions:

• Artificial Intelligence(AI): Technology that uses computation and machines to replicate human reasoning and problem-solving capabilities.
• Machine learning (ML): Algorithms analyze large amounts of data in real-time to find patterns and trends, helping make better decisions.
• Deep learning (DL): A technique that uses complex neural networks that work like the human brain to find important information in unstructured data, such as text, audio, and images.
• Large Language Models (LLMs):  AI systems designed to understand, generate, and manipulate human language with a high degree of proficiency. LLMs are trained on datasets, which are databases of textual data, that allow systems to understand, generate, and manipulate human language.
• Natural language processing (NLP): Technique that allows computers to understand human language in news articles, online comments, and other information to spot events that affect markets and measure investor feelings.
• Predictive analytics: Process of using machine learning, deep learning, data mining, statistical modeling, and other advanced technologies to predict future price changes and market volatility based on past data and current trends.
• Risk management systems: AI algorithms help monitor trading activities and evaluate potential risks continuously.
• Inference: Invoking an LLM or ML model, and accessing compute power to get a response.
• Zero-knowledge Proof (zkP): A technique where one person (the prover) can show another person (the verifier) that a certain statement is true without revealing any other information besides the fact that the statement is true (e.g. generating proofs).
• Zero-Knowledge Machine Learning (zkML): A cryptographic approach that enables the execution of machine learning models on data without exposing the data itself to the model provider. This technique leverages zero-knowledge proofs to ensure privacy and security.
• MPC (Multi-Party Computation): A cryptographic protocol that allows multiple parties to jointly compute a function over their inputs while keeping those inputs private from each other.
• RAG (Retrieval-Augmented Generation): A method where a model retrieves relevant documents or pieces of information from a user-defined database to improve the accuracy and relevance of generated responses.
  • ​​Vector storage: A method of storing and organizing data (documents like CSVs, and PDFs) in the form of vectors, which are numerical representations of data points (for vector search).
  • Vector Search: A search method that uses vectors to find similar items in a dataset by calculating the distance between these vectors.
  • Embeddings: Numerical representations of data (such as words, sentences, or images) in a continuous vector space, used to capture semantic relationships and improve the performance of models.
• AEA (Autonomous Economic Agent): A software agent that can perform economic transactions on behalf of a user autonomously, making decisions and interacting with other agents or systems.
• OEF (Open Economic Framework): An infrastructure that provides the protocols and tools necessary for deploying and operating autonomous economic agents (AEAs) in a decentralized digital economy.

Fetch AI - the L1 for all things on-chain and IoT

Fetch.ai is an L1 network (powered by the Cosmos SDK) for building and deploying AI agents that integrate into legacy systems.

With Fetch.ai, developers can automate business operations tasks like updating records and managing complex workflows. Fetch.ai smart agents also seamlessly integrate with IoT devices, allowing agents to interact with hardware used for industrial services.

The project's founding team brings extensive experience in the AI field. Sheikh, the CEO, provided early-stage funding to DeepMind before its acquisition by Google, and the CSO, Hain, is a professor at the University of Sheffield with expertise in machine learning and AI.

To date, Fetch.ai secured ~$61.9M in funding from investors like DWF Labs, and GDA Group.

Users can access the Fetch Network and its AI-powered apps by downloading the Fetch Wallet and topping it up with FET tokens. The Fetch wallet even allows users to automate position management with a chat interface.

How Does Fetchai Work?

Fetch.ai Agents are programmable bots that can complete tasks, from generating code examples to automating tasks and commanding IOT devices. Fetch.ai provides an in-browser IDE for developers to create agents using Python and JavaScript.

Agents are coordinated within the “Agentverse”, which is a decentralized environment where all agents are created, registered, and connected with other AI agents. Using Fetch.ai’s Open Economic Framework (OEF), agents within the Agentverse can discover work opportunities from other agents, exchange data, and access services.

You can think of the Agentverse as a stock trading floor, and the agents as brokers hard at work. But instead of agents fulfilling public securities orders, their purpose is to interact with the environment and other agents to complete tasks assigned by a user.

Tokenomics

The FET token was launched in February 2019 as an ERC-20 token. With Fetch.ai now having its own L1 blockchain, the FET token is native to the Fetch Network but can still be bridged back and forth to ERC-20 FET.

The Fetch.ai foundation minted 1.48 billion FET tokens, on May 3rd, 2024, in anticipation of a token merger with SingularityNet (AGIX), and Ocean Protocol (OCEAN). This increased the total supply from 1.2b FET tokens to 2.6b FET tokens ahead of the transition to their combined token (ASI).

Total Supply: 2,630,547,141 FET

• Foundation: 20.0% (2.75% at TGE, 47-month vesting)
• Founders: 20.0% (6.25% at TGE, 33-month vesting)
• Token Sale (Seed, Private, and Public Sale): 17.6%
  • Public Sale: 6% (100% at TGE)
  • Seed: 5.24% (3-month cliff followed by 2-month vesting)
  • Private: 6.38% (3-month cliff followed by 2-month vesting)
• Future Releases: 17.4% (1-year cliff followed by 59-month vesting)
• Mining: 15.0% (6-month cliff followed by 7-year vesting)
• Advisors: 10.0% (12.5% at TGE, 33-month vesting)

Current Token Stats

• Circulating Supply: 2,521,012,371 FET tokens (95.8%)
• Market Cap: $3,908,223,058
• Fully diluted valuation (FDV): $4,078,030,362

FET Token Utility

• Governance: FET holders can vote on governance proposals and decisions that affect the development and direction of the project.
• Staking: FET holders can stake their tokens to secure the network and earn rewards. Staking also gives them access to additional features such as collective learning.
• Network Fees: FET is used to pay for the fees associated with using the platform's services such as creating agents, accessing data, executing smart contracts, etc.
• Incentives: FET is used to incentivize users and developers to contribute to the platform's growth and innovation. For example, users can earn FET by providing data or liquidity to the platform, while developers can earn FET by creating applications or services for the platform.

Note: The scope of the utility will change once FET transitions to the ASI token

Our take on Fetch.ai

The token is almost fully vested and “Mining Rewards” (for nodes) and “Future Releases” are the only allocations left to be unlocked, so it seems there's little to be worried about in terms of sell pressure.

However, recently in May, there was an increase in market cap, which was not reflected in token price, which means more tokens were minted. And then both market cap and price experienced a massive drop, which means those tokens hit the market…hard. The price of FET dropped ~51%, from May 22nd to June 19th, 2024.

This is because the Fetch.ai foundation minted an additional 1.48 billion $FET tokens, (~$2.96 billion), to support the ASI Token Alliance on May 3rd, 2024.

The ASI token alliance merger is a plan to consolidate SingularityNET’s AGIX token and Ocean Protocol’s OCEAN token into Fetch.ai’s FET token, before transitioning to the ASI (Artificial Superintelligence) token.

The migration will take place over 2 phases where holders of previous tokens can exchange them for FET at the following rates:

• $AGIX tokens migrate to FET, at a conversion rate of 1 $AGIX to 0.433350 FET
• $OCEAN tokens migrate to FET, at a conversion rate of 1 $OCEAN to 0.433226 FET

‍

So, what does this mean for degens?

To simplify things: SingularityNet and Ocean Protocol are working on some vital infrastructure that could make these partnerships worthwhile.

Ocean is a decentralized data exchange protocol offering tools for Data NFTs & data tokens as well as a marketplace for buying & selling private data.

SingularityNet allows users to create, edit, and manage AI-related services like model training and inference to deploy AI agents. The SingularityNet team is also developing AI-DSL (Artificial Intelligence Domain-Specific Language), a specialized programming language for communication and interoperability between AI agents

Aside from the ASI merger, Fetch.ai could take off if it begins to gain adoption in healthcare, automotive, supply chain management, and digital identities.

At the moment, Fetch is mainly gaining usage from developers looking to enhance internal operations through automation and “micro agents.” But, imagine if a store owner uses Fetch.ai to set up an agent to watch inventory supply levels, warn the manager when supplies are low, and order new inventory upon request.

One company that sees this potential is Deutsche Telekom (DT), one of the leading telecommunications companies in Germany, which recently became a validator and announced a partnership with Fetch.ai.

Dirk Röder, Head of Web3 Infrastructure & Solutions at DT, stated: “Autonomous agents will automate industrial services, simplify processes, and make them secure and scalable thanks to blockchain technology.”

The future of Fetch.ai and the ASI token seems bright, but I don’t blame you if you don’t want to pick up FET in the meantime.  Since the tokenomics are under construction for the foreseeable future, just about anything can happen to the price - as we’ve seen with the ~50% drop in FET price from May to June.

Autonolas - plug-and-play agent building

Autonolas is creating a network of general-purpose AI agents with plug-and-play components. DAOs can tap into Autonolas’ infrastructure to build agent-powered services, and developers can contribute code to Autonolas’ protocol-owned agents for compensation.

Autonolas conducted an IDO on Fjord Foundry, raising ~$540k, between July 5th to July 12th, 2023. Valory, the team behind Autonolas, also raised ~$4m in its seed round, from investors including True Ventures and Signature Ventures.

The protocol is live on 8 public blockchains including Solana, Ethereum, Arbitrum, and Gnosis Chain.

How Does Autonolas Work?

Autonolas structures agents into AI agent services, where a group of independent computer programs interacts with each other, sharing one application state and logic.

This approach is made possible by the autonomous economic agent (AEA) framework.

The AEA framework allows builders to create agents that:

• Communicate in a peer-to-peer manner.
• Interact directly with blockchains (such as Ethereum) and their smart contracts.
• Implement arbitrary business logic for reuse through components.

Each agent comprises plug-and-play components that developers can use to create their own agents. Developers can combine existing components or develop their own components to create agents for specific needs.

Let’s say a DAO creates an agent service that automates treasury management, and publishes it to the marketplace. Other developers can choose to leverage the unique set of contracts the agent services’ interacted with, or the logic the smart agents used to complete the task.

The open-AEA framework allows for the following reusable components:

• Protocols: peer-to-peer communication. Defining agent-to-agent interactions and component-to-component interactions within agents.
• Connections: interfaces for the agent to connect with other agents and external APIs.  
• Contracts: wrap logic to interact directly with smart contracts of decentralized apps.
• Skills: self-contained capabilities that the agent can adapt as needed in different situations.

At the moment Autonolas does not charge fees to use its infrastructure. Instead, the protocol accepts donations based on the usefulness of its poSe and infrastructure. Autonolas mainly directs donations towards compensation for developers, and the growth of Protocol-owned Liquidity (POL).

Autonolos earns donations in two ways:

• When a DAO uses Autonolas infrastructure they can make donations to the protocol (and authors of agent components that were used).
• Autonolas poSe services receive donations from community members. The protocol then distributes donations to developers (Component authors) and the protocol treasury (including POL).

If you’re a developer, you can create agent services and components for Autonolas’ DAO, Protocol Owned Services (PoSe).

To track royalties on-chain, Autonolas issues NFTs representing code and services. This way, rewards can be attributed to the developers.

Example Scenario

DAOs can use Autonolas’ infrastructure to create and monetize agent services.

As outlined in the project whitepaper, a DAO can create a hiring recommendation system using Autonolas Agent services. To cut the time it takes to find qualified candidates for open positions, employers would contact the DAO and use this agent service.

In this case, the DAO is the service owner. The service owner earns revenue by charging “service users” —individuals using the service agent to find a job.

The Service owner can choose to give the protocol a donation, which Autonolas would use to compensate developers of agent components and line the protocol treasury.

Tokenomics

Total Supply: 538,674,107 OLAS

• DAO Treasury: 10%
• Valory Team allocation: 10%
• DAO founding members: 32.65%
• Rewards:  47.35% (2% inflation after 10 years) - Developer and bonding rewards

Current Token Stats

• Circulating Supply: 41,512,652 OLAS tokens (7.7%)
• Market Cap: $69,728,401
• Fully diluted valuation (FDV): $904,808,310

OLAS token Utility

• veOLAS: Locked OLAS (Users can lock OLAS or bond to participate in governance to modify system parameters, add new functionality to the on-chain protocol, and more). veOLAS can vote on parameters that determine how donations accrued by the Protocol, and newly minted OLAS are shared to developers and the treasury.
• Burnable OLAS (buOLAS): Tokens connected to team vesting for the founding DAO members; the majority of which is set to unlock in mid-to-late 2026.
• Valory: Working capital for the team behind AutonOlas.
• DAO Treasury: Treasury for protocol initiatives.

Autonolas uses a bonding mechanism similar to Olympus DAO, where holders of OLAS can provide liquidity to DEX and sell their liquidity provider (LP) tokens (i.e. OLAS-DAI, OLAS-USDC, or OLAS-ETH) to the protocol for OLAS at a discounted price.

This creates a source of funding for the protocol by stimulating liquidity for OLAS, consistently. Furthermore, it increases protocol-owned liquidity (POL), ultimately increasing the value of Autonolas. POL represents the protocol assets under management, including some treasury funds and LP positions.

Autonolas introduces a control mechanism to ensure that OLAS is not excessively bonded or locked away from circulation. This prevents excessive concentration of power in the hands of a few users.

Our take on Autonolas

The Good Side

~44% of OLAS is locked as veOLAS and the majority of veOLAS is locked for the max duration (4 years).

The protocol holds ~$15m in POL, consisting of various OLAS LP pairs (i.e. OLAS-USDC, OLAS-DAI, OLAS-ETH, etc.) across several blockchains, including, Ethereum, Arbitrum, and Base.

The protocol contributes to its POL through this bonding as well as donations when its protocol-owned services (poSe) are used.

Autonomous Service Activity has picked up, as shown by total weekly transactions rising from 5,980 in November 2023, to well above 30,000 TXNs per week starting in March 2024.

Developers are slowly starting to adopt Autonolas tech stack, and they’ve recently reached 554 complete services, and 800 components for customizing smart agent services:

Most transactions are occurring on Gnosis Chain and Polygon:

The protocol has also received ~27.5 ETH in donations (revenue from poSe) since inception.

The Bad Side

Insiders hold ~42% of the supply, and the DAO treasury funds haven’t been touched, which means two things: (1) Insiders control an uncomfortable amount of tokens, and (2) There are still a lot of funds to be deployed by the DAO, Valory, and other insiders – meaning there’s a lot of sell pressure left for OLAS.

With OLAS’ low circulating supply, the ve-mechanics of veOLAS create uncertainty around prices during unlocks.

The project is steadily gaining interest, showing decent potential, but, at the moment, it lacks the interest necessary for sustainable revenue long term.

Autonolas is currently using a donation model to crowdsource developers to create useful projects that direct value (through donations) back to the protocol. It is likely that Autonolas is using this strategy to attract developers and will eventually switch to a fee-based model for using the infrastructure.

Regardless of the business model, if Autonolas doesn't attract developers to build with its tech stack faster than other agent provider protocols, the project may be out of luck.

Morpheus - the future of crypto UX?

Morpheus provides general-purpose “smart agents” that connect to user wallets and interact with smart contracts. Morpheus’ smart agents can fetch on-chain data, buy or sell crypto, perform position management tasks, and more.

The platform features an LLM trained on Web3 data including Blockchains, Wallets, Dapps, DAOs, & Smart Contracts. Users can store long-term memory of their data & connected applications – to provide a broader context to Smart Agent actions– locally or via a decentralized cloud.

Morpheus also features an Agent App Store allowing developers and node operators to contribute code and compute power to receive MOR rewards.

How Does Morpheus Work?

Morpheus employs Smart agents, which are on-chain AI agents capable of interacting with smart contracts and fetching blockchain data.

Morpheus’ ecosystem consists of four key contributors:

• Community - Builders create tools and infrastructure to bring users and developers into the Morpheus ecosystem.
• Capital - Users that stake stETH and direct rewards to Morpheus in exchange for MOR – participating in the “Techno Capital Machine”.
• Compute - Provides the equipment and compute power for inference LLMs. The user providing compute must run a full node, hold a MOR address, and stake MOR.
• Coders - Developers that work on the Morpheus codebase.

Example

If you want to use Morpheus to swap tokens: download and install a smart agent app, connect your Web3 browser wallet, and purchase MOR on Arbitrum.

Let’s say you want to swap ETH for WBTC, you can use the app's chat interface and tell the smart agent “Swap ETH for WBTC”. The agent will then prompt you to specify the amount.

Once all necessary information is gathered, the agent will look up the assets on the current chain, retrieve contract addresses, and generate a quote for you to review.

If you accept the quote, the back end will create the necessary transactions and send them to you to sign using your wallet.

MOR Tokenomics

Approximately 300,000 MOR tokens (0.71% of the total supply) were minted for a Morpheus’ fair lunch event, where users staked stETH, and directed their yields to Morpheus for MOR tokens, on February 8th, 2024.

Total Supply: 42,000,000 MOR

• Coders: 24% (3,456 tokens/day - 16-year vesting)
• Capital: 24% (3,456 tokens/day - 16-year vesting)
• Compute:  24% (3,456 tokens/day - 16-year vesting)
• Community: 24% (3,456 tokens/day - 16-year vesting)
• Protection funds: 4%

Note: The block reward will start at 14,400 MOR per day and then decline by 2.468994701 MOR each day until the reward reaches 0 on day 5,833.

Current Token Stats

• Circulating Supply: 330,589 MOR tokens (0.78% of the total supply)
• Market Cap: $9.73M
• Fully diluted valuation (FDV): $1.24B

MOR token Utility

• Users: Users can pay MOR for specialized Agents released by developers.
• Capital: Capital providers receive MOR rewards prorated to stETH they contributed against all stETH deposited. Capital providers equip developers with liquid funds and line Morpheus’ protocol-owned liquidity (POL).
• Compute:  Compute providers receive MOR for the user prompts they reply to.
• Coders:  Coders receive MOR for the merged contributions they make to the Morpheus codebase.
• Community: Community builders receive MOR for the front ends, tools, usage, & value they bring into the Morpheus network. stETH yields from the community are directed to Uniswap's liquidity pool, which increases daily.

Our take on Morpheus

The protocol is doing a good job of ensuring long-term success by incentivizing proof of work (compute providers and coders) and proof of stake (stETH stakers) over 16 years.

Developers are leveraging Morpheus to build projects like 6079ai, a security protocol for trustless & verifiable GPU compute – LLM & ML inferences, and Venice, a protocol for private and uncensored AI chatbots with RAG capabilities.

Additionally, for users, Morpheus seems like a cool way to automate on-chain workflows for DAO governance, and simple yield farming transactions using a chat interface.

The FDV is 127x more than the market cap, but MOR has a bitcoin-esque design so the low float isn’t too worrisome if Morpheus community members continue to hold MOR or use it for services within the ecosystem.

Currently, “Capital” providers have deposited ~$272m worth of stETH, to catch a 12% yield in MOR, from Morpheus protocol. We’ll have to see what they do with tokens going forward.

Morpheus’ whitepaper proposes a tail emission that will take place in the future to burn tokens and take them out of circulation. Burns will be calculated by reviewing the number of burned MOR tokens in the past 5,833 days and the tail emission value will be set to 50% of the burned amount.

If Morpheus implements tail emissions and manages to gain more users willing to pay for community smart agents, MOR might be attractive in the near future (1-2 years).

For a deeper dive into Morpheus, check out our previous meal deal report.

Spectral - accelerating code deployment

Spectral is a protocol providing AI agents that can write and deploy Solidity code. Spectral Labs received ~$30M in funding from Investors including Franklin Templeton, Polychain Capital, Google, and more.

Spectral’s “Syntax” product lets users easily create on-chain agents from simple instructions that leverage machine learning workflows via Spectral “Nova”.

Spectral agents can manage smart contracts and assets based on programmable logic to achieve things like creating and launching tokens, deploying novel contracts like DAO voting platforms, and more.

Syntax features a “Co-pilot” (chat interface) that allows you to spin up complex contracts in a matter of minutes. The co-pilot takes natural language inputs, similar to chatGPT, and converts them into code-based Agent instructions.

If you’re a smart contract developer, you can tap into your Github repo to make changes to existing projects. The platform also features a Retrieval-Augmented Generation (RAG) system to enhance responses using protocol documentation, code examples, and other content.

Spectral ensures secure smart contract development with Foundry, an open-source toolkit for Ethereum development that compiles, tests, and deploys all smart agent contracts.  

How does Spectral Work?

Spectral is deployed on Arbitrum, and leverages the network for:

• Storage for inference requests, model challenges, and fraud proofs for Solvers, Challenges, and Validators.
• Registration/tracking for all the actors (Solvers, Consumers, Solvers, and Creators).
• Contracts for interacting with Spectral Network (for committing models, consuming inferences, etc.).

The team developed an ERC4337-compliant multi-signature wallet for their machine-learning platform, enabling gasless transactions on Arbitrum.

The platform consists of three main components, including:

• Spectral Syntax: A network of Onchain Agents. Syntax allows users to create their own on-chain Agents through a Solidity co-pilot. Includes a custom-built dataset comprising over 7 million Solidity contracts, to produce accurate and gas-optimized Solidity code.
• Spectral Nova:  A machine intelligence network that provides decentralized machine learning inferences directly to smart contracts. These inferences can be verified via verifiable computation (e.g. zero-knowledge machine learning, optimistic machine learning, etc.).
• Interchain: The layer interconnects Syntax and Nova, facilitating communication between on-chain AI Agents and ML inferences.

Nova incentivizes data scientists and ML engineers to create models that produce inference feeds for solving predictive and machine intelligence problems. This allows users to access the necessary inference feeds directly.

These inferences are used to power contracts on Syntax, and Inferchain ensures seamless communication between these agents.

The platform uses verifiable computation techniques to ensure inference integrity. Zero-Knowledge Machine Learning (zkML) mathematically verifies inferences without revealing the model, while Optimistic Machine Learning (opML) efficiently verifies inferences on-chain, generating proofs only when an inference is challenged.

Spectral is also partnered with a few Web3 giants including IPFS, for proof storage and inference submission, and Alchemy as an RPC node provider to interact with Arbitrum.

These approaches guarantee that users can trust the accuracy and origin of the inferences.

SPEC Tokenomics

8% of the total supply (8,000,000 SPEC) was airdropped to early adopters of Spectral including developers, and community members on May 26th, 2024.

Total Supply: 100,000,000 SPEC

• Community: 40.0% (12-month cliff, followed by 30-month vesting)
• Core Contributor: 22.0% (12-month cliff, followed by 30-month vesting)
• Investors: 38% (12-month cliff, followed by 18-month vesting)

Current Token Stats

• Circulating Supply:  10,523,351 tokens (10.5% of the total supply = airdrop + liquidity allocations)
• Market Cap: $66,776,857
• Fully diluted valuation (FDV): $634,558,835

SPEC Token Utility

• Solvers: Stake SPEC tokens to join machine learning challenges and earn rewards based on their models' quality and performance. Users pay fees in Ethereum (ETH) or stablecoins to access these models, with a portion going to Solvers.
• Validators: Stake SPEC tokens to verify Solvers' submissions and ensure they follow challenge rules, earning rewards from inference requests and for exposing misconduct.
• Consumers: Pay fees in SPEC, ETH, or stablecoins to access machine learning models, with fees going to Solvers. Staking SPEC gives consumers discounts or fee waivers. SPEC also serves as a governance token for voting on network upgrades and securing the network, offering benefits like lower fees, faster transactions, and access to exclusive Agents.
• Creators: Post data science/ML challenges, set performance benchmarks, and earn revenue from the use of their inferences generated from their challenges.

Our take on Spectral

Insiders control 60% of the total supply, but the project allocated the remaining 40% to the community and has shown a commitment to developers and community members by airdropping ~8% of the total supply.

There are also still a lot of tokens to hit the market, and the supply won't be fully unlocked for ~3 years.

Spectral’s use case of smart agents that write code is straightforward and the project is backed by well-known investors, which should help it get good media attention if it meets its goals.

However, it needs developers and tech-savvy farmers to trust and use its Spectral agents for debugging and deploying these contracts. Smart contracts control billions of dollars, and developers can’t afford to risk a hack because they tried to cut corners with AI.

The Spectral Labs CEO Sishir Varghese also recognized the apprehension in trusting smart agents that code: "I think skilled developers are still pretty apprehensive about all of the copilot stuff – except in Web2. For Solidity, this is going to be a brand-new experience, to be honest."

We'll have to keep an eye on this project. The SPEC token could become attractive in the next 1-2 years as more tokens hit the market. But the success of their SPEC token will ultimately rely a lot on how many developers start using their agents and how reliable Spectral’s code proves to be.

Talus - a next-gen L1 for smart agents

Talus is an L1 blockchain for designing and deploying on-chain smart agents. The network is powered by the Cosmos SDK and employs Sui Move (and the move VM) as the native programming language.

According to Talus’ Litepaper, developers can build agents that manage yield strategies, in-game resource collection, IoT workflows, and more. Talus also supports Initial Model Offerings (IMOs) so projects can collect funds and computational resources for community development.

Talus raised $3M in funding led by Polychain Capital, with the participation of dao5, Hash3, and others.

The L1 network integrates Move's smart contract security and performance, empowering developers with robust tools for AI smart agent applications.

The MoveVM enhances application security and performance, resolving common vulnerabilities before code deployment.

How Does Talus Work?

Talus’ protocol consists of the following components:

1. Protocol (consensus & execution)
   1. Protochain Node (Cosmos SDK L1)
      1. IBC (Inter-Blockchain Communication protocol via Cosmos SDK)
   2. MoveVM (run time for Sui Move programming language)
2. Mirror Objects (off-chain objects like LLM inferences, oracles, and datasets)
3. AI Stack (SDK for developers)
4. Smart Agents

Protocol provides the consensus, execution, and interoperability foundation for building onchain smart agents

Protochain Node

Protochain is Talus’ Proof-of-Stake (PoS) “blockchain node”, powered by Cosmos SDK and CometBFT.

By using the Cosmos SDK, Talus protocol can leverage the Inter-Blockchain Communication (IBC) protocol, which enhances the capabilities of its smart agents in a few ways:

• Interoperability: facilitates seamless interoperability between different IBC-enabled blockchains, enabling smart agents to interact and utilize data or assets across multiple chains.
• Atomic Transactions: IBC supports atomic transactions across chains, ensuring operational consistency and reliability, crucial for complex financial applications.
• Scalability: by enabling smart agents to operate across multiple blockchains, IBC indirectly supports scalability through sharding, and distributing transaction processing load effectively.

IBC ensures security and isolation between chains, enabling smart agents to operate confidently across chains without compromising security.

MoveVM (Sui Move)

Virtual machines (VMs) enable the execution of smart contracts by defining transactions and how blocks are created. They often work with specific programming languages.

The MoveVM is designed for the Move and Sui Move programming languages.

The Move programming language, along with MoveVM, boosts on-chain security by fixing critical issues found in Solidity, such as re-entrancy and access control flaws.

This makes it easier for developers to create secure protocols for managing valuable resources on the blockchain. Move also simplifies the creation, transfer, and management of digital assets by treating objects as important components.

The Move programming language supports native ownership and efficient handling of assets, which is essential for implementing AI resources directly. MoveVM, ensures high performance through parallel processing, allowing the network to manage multiple tasks at once.

Mirror Objects

Developers can use Mirror Objects to represent off-chain resources (services) like AI and machine learning inferences, as well as datasets, on chain.

Since complex AI computations can't be done directly on the blockchain, Talus coordinates the exchange and verification of these transactions via Mirror objects. Essentially, Mirror Objects bring off-chain resources into Talus’ on-chain ecosystem.

These objects also provide key functions such as proving the uniqueness of resources, enabling their trade, and verifying ownership.

There are few types of Mirror Objects:

• Model Object: Model owners can add their AI models to the ecosystem via a model registry, creating a Model Object. This digital representation includes the model’s features and capabilities and frameworks for ownership, management, and monetization.
• Data Object: A Data Object, or Dataset Object, is a unique dataset. Creating a Data Object involves committing a dataset and proving its storage. Talus’ verification contracts check the registry to confirm the data’s uniqueness, and establish the object.
• Computation Object: A verifiable computation program (Inference). A buyer (developer) proposes a computational task to the hardware owner, who’s required to deliver the result and proof. The buyer then decrypts the commitment to verify the result. If validated, the transaction moves forward.

Smart agents on Talus can offer different verifiability capabilities suited for the specific needs of developers (for AI model inferences). Here are some possible approaches for developers, based on their potential concerns:

AI Stack

Talus’ AI stack offers a software developer kit (SDK) and integration components for developers to build apps powered by Talus smart agents.

The SDK includes:

• Libraries that work across different systems and programming languages.
• Data models and smart contracts (building blocks for smart agents).
• Oracle integration.

Integration components provide an efficient way for tasks like interfacing with Oracles. These components also serve as building blocks that developers can leverage for future integrations.

Smart Agents

Talus' protocol (Protochain Node & MoveVM) provides a decentralized environment for smart agents, enabling them to interact with other agents and humans to complete tasks and achieve goals.

These agents can also communicate with each other, allowing for more complex behaviors.

Based on programmed logic, agents can take initiative based on their goals, predictions, or future expectations, and they can pursue set objectives, optimizing transactions for factors like energy efficiency or adjusting network settings.

Talus Tokenomics

Talus Network hasn’t launched its TAI token, yet, but it has detailed a brief tokenomics plan in its whitepaper.

When the token is live, TAI will serve as the sole currency for all transactions, including:

• Smart agent usage
• The purchase and trade of resources (computing, data, or models)
• Network staking for security consensus
• Governance

Talus Network is currently running a private testnet. If you’re interested, you can sign up on the Talus Network website.

Our take on Talus

Solana and Ethereum developers that have transitioned to Move-based chains like Aptos and Sui express decent praise for the Move programming language’s ease of use and built-in safety features.

Talus’ implementation of the Cosmos SDK allows it to communicate with other blockchain networks in the Cosmos ecosystem, and leverage features like atomic swaps.

If Talus attracts developers to help build an L1 for smart agent-powered apps, the token could gain a lot of attention.  

We’ll have to keep an eye on this project and its development, especially with projects like Morpheus already having a live product where users can automate simple yield strategies with a chat interface.

Shinkai - your personal AI side-kick

Shinkai is an AI operating system for managing everyday tasks, from content writing and editing to fetching on-chain data.

When you download Shinkai on your desktop, it establishes a Vector File System (VectorFS), allowing you to import your data for a more personalized experience. This means you can spin up a context-specific AI financial advisor or a personal assistant by providing Shinkai access to your spreadsheets, PDF files, and other documents.

Additionally, Shinkai agents are capable of tasks like controlling multi-sig wallets and interacting with DAOs and on-chain protocols.

How does Shinkai work?

At Shinkai’s core is the Shinkai Node, which runs on user CPUs and acts as a personal gateway between users and Shinkai’s network (after a Shinkai Identity is created). Under the hood, nodes manage tasks such as agent management, job processing, payment processing, and secure communications between other nodes.

The platform employs a Zero-Knowledge Multi-Party Computation (MPC) Data Protocol to ensure data authenticity without compromising user privacy.

Since the Shinkai node is written in Rust, it can leverage a zkVM (zero-knowledge virtual machine) known as the Succinct Processor 1 (SP1), to generate proofs.  

Zero-knowledge proofs in a Multi-Party Computation protocol assure the authenticity of data provided to AI without compromising privacy.

Furthermore, the network's peer-to-peer structure supports real-time secure AI data sharing and communication, which is further enhanced by router/proxy nodes for better scalability and privacy.

For fine-tuning, Shinkai comes out of the box with Vector File Storage (VectorFS), a file management system that stores your data as embeddings, with full Vector Search capabilities.

Vector Search is the technology that allows LLMs to find and use needed information from data sources at scale, which is known as RAG (retrieval augmented generation). Simply put, you can store your PDFs and CSVs so the AI can have more personalized context.  

Shinkai Desktop runs locally as an app, but if you’d like to connect to the shinkAI network, you’ll need to stake KAI tokens.  

Connecting your Shinkai node to the network allows you to access up-to-date data for cryptocurrencies, news events, merged push requests on Github, and more from external sources to supercharge your agent’s abilities.

After staking KAI tokens, your stake accrues inflation awards, and you receive a Shinkai Identity, an on-chain NFT, serving as your unique ID. Your Shinkai Identity connects directly with your Shinkai node, allowing all messages, files, or anything else anyone (or any AI Agent) sends to you to seamlessly arrive at your node.

Representing IDs as NFTs grant Shinkai Identity a few characteristics that increase interoperability for users, including:

• Ownership and Transferability: Shinkai Identity NFTs can transfer ownership of the identity (and thus associated stake) to other addresses.
• Infinite Sub-identities: Underneath a single Shinkai identity, an infinite number of sub-identities can be created for all devices/computers/AI agents that the user owns.
• Router/Proxy Nodes: By specifying one or more router/proxy nodes the network supports further scalability, enhanced message delivery assurance, and privacy.

Shinkai Tokenomics

Shinkai is still in its very early stages. In anticipation of Shinkai’s mainnet and the KAI token launch, the team is hosting the first of several incentivized testnet rounds on CoinList. The Shinkai Testnet is split into three phases:

• Phase 1: June 6, 2024 - June 27, 2024
• Phase 2: July 2, 2024 - July 23, 2024
• Phase 3: July 30, 2024 - August 20, 2024

The team is allocating 24,000,000 KAI tokens (2% of the total supply) to reward users of all rounds.

If you’re interested in participating you can join the Galaxe campaign. You’ll be required to try the Beta, complete certain tasks, and provide feedback. In return, you’ll qualify for the KAI airdrop.

Besides the Galaxe campaign, Shinkai is incentivizing usage in other ways:

• Developers: Rewards up to 20,000 KAI tokens based on the quality and usefulness of their tools and workflows.
• Content Contributors: Rewards up to 20,000 KAI tokens based on the quality of their content, the interest it generates, and the number of subscriptions it attracts.
• High-Quality Feedback: Rewards based on the quality of feedback, especially for detailed bug reports and suggestions for UX improvements and new features. Points are given based on the severity of the issues reported and whether they are new.

Our take on Shinkai

Shinkai is entering the market for personalized chat assistants with RAG-capabilities, which puts it up against other open-source (non-crypto) projects like Langflow, and FlowWise. Shinkai’s competitive edge is its price point (value of tokens required for staking), and crypto knowledge base.

If you have some free time, you should join the Incentivized testnet, especially if you want to qualify for the airdrop, and try out the Beta to see how Shinkai can fit into your workflows.

Right now, only about 7.3k users are in the Galaxe quest, which isn’t a lot. This gives you a chance to join an under-farmed airdrop campaign and try out a new on-chain chat assistant and agent with RAG capabilities.

Wayfinder - pathfinding through the cryptoverse

Wayfinder is focused on “omni–chain” AI agents that interact with smart contracts and navigate across blockchains to complete onchain tasks for users. The protocol will initially launch on the Solana network but aims to work across the majority of blockchain ecosystems, in the future.

The Wayfinder app offers a chat interface for users to craft prompts for their agents, with the use of various elements like wayfinding path information, price data, memories, cloning and ownership management, and uploaded content.

The protocol was originally developed to serve as the core infrastructure for Colony, Parallel Studios’ upcoming strategic survival simulation game. It’s intended for Colony players to use Wayfinder shells (AI agents) as a way to harvest and manage rewards for in-game workflows.

Parallel Studios, the team behind Wayfinder, received ~$85M in funding from investors including Paradigm.

Wayfinder's ultimate goal is to create a sophisticated network of paths and leverage machine learning so agents can develop logic to choose the correct paths for specific tasks based on user preferences.

How Does Wayfinder Work?

Wayfinder's core components include:

• Wayfinding Paths - Edges that connect between nodes to create paths between one smart contract and another
• Shells - AI agents (manage wallets and private keys)
• Wayfinding Graph - Ecosystem graph of smart contracts represented as nodes

Wayfinder’s architecture is based on wayfinding paths, a community-developed knowledge base of viable paths for shells.

Zooming out, the paths reveal a map of nodes represented as blockchain-based destinations (smart contracts), known as the ecosystem graph. Using the ecosystem graph, AI agents can efficiently navigate blockchains, producing “breadcrumbs” along the way to help other agents use the same paths in the future.

Simply put, when shells complete a task, they index the blockchain destinations (smart contracts) visited, so other shells can learn from and use the same paths.

Nodes in the Wayfinder graph can represent:

• Protocols: Encompass applications, contracts, and assets tied to a specific protocol, providing essential information about their functionalities and usage.
• Contracts: Represent on-chain logic and applications within a protocol, each identified by a unique address, detailing their specific functions and interactions within the blockchain ecosystem.
• Contract Standards: Include nodes like ERC-20, ERC-721, ERC-1155, and SPL libraries, defining the functionalities of contracts and assets, crucial for understanding their utilization and interactions.
• Assets: Comprise tokens and NFTs, capturing key details such as ticker symbols, contract addresses, and implemented standards, offering a comprehensive overview of their characteristics and roles.
• Functions: Segments of code that perform specific blockchain actions, ranging from individual tasks to protocol-level operations like executing swaps, and enabling direct blockchain interactions.
• API Functions: Protocol-level functions that access off-chain data, essential for protocols relying heavily on external information.
• Routines: Sequences of functions structured to complete specific tasks, outlining necessary steps and functions to facilitate complex blockchain operations.

With the options provided by the ecosystem graph, shells can evaluate competing priorities, and choose a path that completes the task most efficiently. For instance, one user may instruct their shell to choose the most established paths, while another user may prioritize lower gas fees.

Tokenomics

Wayfinder hasn’t released its token yet, but it detailed plans for its upcoming PROMPT token, in the project whitepaper.

Total Supply: 1,000,000,000 PROMPT

• Launch Partner Treasury: 1.34%
• Investors: 25.39%
• Foundation Treasury: 6.66%
• Team: 16.51%
• Community
  • Wayfinding Rewards: 5%
  • Future Incentives: 5%
  • Cached: 40%

Current Token Stats

• Circulating Supply: N/A
• Market Cap: N/A
• Fully diluted valuation (FDV): N/A

PROMPT token Utility

When the PROMPT token is live, users will require PROMPT to create new shells that perform tasks. The token could also be used to reward developer bounties.

Although the token isn’t live yet, if you have PRIME, Echelon Prime’s governance token, you can “cache” it to receive rewards in a locked PROMPT. To cache your PRIME, visit Wayfinders website and log in.

Echelon Prime provides tooling for developers to create games that integrate with blockchain technology. Wayfinder wants the Echelon Prime community to control the majority of governance when the protocol launches initially.

Our take on Wayfinder

The project has solid use cases, like a Colony AI gaming assistant, and on-chain companions, which could push adoption and help the PROMPT token accrue value.

Wayfinder is also demonstrating a commitment to the community, with over 50% of the total supply being allocated to the community.

Picking up some PRIME seems like a decent play, but I don’t know if I’d deploy funds to PRIME just to cache it for PROMPT. There are way too many other attractive options in the market at the moment.

The only thing that kind of turns me off from PRIME is that Wayfinder will be geolocating individuals interacting with the network, and will incorporate government-published lists of sanctioned digital addresses. As stated in the whitepaper, shells will also inherit any restrictions or regulatory obligations applicable to users of a certain “nationality, residence, or other applicable criteria.”

Other than that, Wayfinder appears to be a solid project, and the PROMPT token allocation is a step in the right direction. We’ll have to keep our eye on the Wayfinder team as they release a beta and develop their tokenomics more.

Oraichain - the oracle for AI APIs

Oraichain is an oracle for AI APIs with a framework for quality-assuring AI models. Under the hood, Oraichain is an L1 blockchain based on Cosmos SDK using Tendermint’s BFT consensus.

With Oraichain, developers can access a diverse marketplace of algorithms and models and leverage computing power to develop AI-powered apps, like automated yield platforms. Developers can also publish their work on Oraichain and earn rewards from users.

Users can access the Oraichain network and its AI-powered dApps by downloading the OWallet and visiting Oraichain Homebase.

How does Oraichain work?

When developers build AI-powered apps they perform a request to call an Oracle Script hosted on the ORAI gateway (marketplace).

Oracle Scripts contain AI data sources (provided by AI providers), test cases, test sources (optional), and transaction fees for each request. Validators are chosen at random to perform this request and validate the data.

The chosen validator fetches data from one or more AI providers on behalf of the user after executing the test scenarios, and if the AI provider fails in testing, the request is canceled.

Validators on the Orai network are required to stake ORAI tokens to secure the network. Similar to other L1s, ORAI token holders delegate their stake to a validator to receive rewards from the network. Successful requests are written to the Oraichain blockchain.

To scale up performance on mainnet, Oraichain employs subnetworks (L2 rollups). Oraichain Subnetworks combines optimistic rollups and zk-rollups to create a novel architecture for its subnetworks, with Proof of Correctness and Proof of Execution mechanisms.

Subnetworks are set to stay in Layer 2 while helping to expand the capacity of Mainnet but not affecting its core logic and security.

At the moment, Oraichain has launched two subnetworks:

• VRF Subnetwork
• AI Execution Subnetwork

Users can download the Owallet Web3 browser wallet and visit the Oraichain Homebase to access apps and features, including:

• OraiDEX: A DEX offering swaps, order books, and futures, based on CosmWasm smart contracts and supporting multichain interoperability (via IBC).
• DeFi Lens: Market insights dashboards powered by generative AI.
• LLM Layer: DeFi chatbot trained on Oraichain with access to on-chain information.
• aiRight: Free AI generative art models and hosting.
• Orchai: Liquid staking provider for scORAI ORAI. Stake ORAI for scORAI for 13.89% (current APY).
• GPU Staking: Stake scORAI to earn GPU demand and block rewards (in USDC).

Oraichain Tokenomics

Oraichain burned ORAI tokens on 3 occasions:

• 73% of initial total supply - December 22, 2020
• 258,449 ORAI burnt from Team - March 28, 2021
• 3,000,000 ORAI tokens from Team, Advisors, and Foundation - June 23, 2021

Total Supply: 19,779,272 ORAI

• Seed Sale: 0.65%
• Marketing: 6.12%
• Ecosystem: 37.54%
• Team: 25.27%
• Advisors: 12.63%
• Foundation: 32.95%

Current Token Stats

• Circulating Supply: 16,144,611 ORAI tokens (81% of the total supply)
• Market Cap: $154,551,248
• Fully diluted valuation (FDV): $171,757,577

ORAI token Utility

The ORAI token is used for the following activities:

• Transaction fees: The ORAI token is required to send requests to the Oraichain network and run any transaction.
• Staking: All validators are required to stake ORAI to be selected to create a block or fulfill data requests. Holders can delegate ORAI to Oraichain nodes.
• GPU-staking: Stake scORAI (liquid staked ORAI) to earn GPU demand and block rewards (in USDC).
• Governance: ORAI token holders can vote on governance proposals like protocol upgrades and parameter changes.

Our take on Oraichain

Oraichain presents an easy way for developers to leverage open-source LLM inferences for apps via the Orai Gateway.

Like any L1, Oraichain is going to have to fight for the attention and capital of on-chain users. The network is home to 30+ AI-powered dApps, which shows the network is building out a solid foundation.

Oraichain stands out amongst similarly sized projects (market cap and TVL).

Moonriver, an L2 network, has a market cap of $104m, which is comparable to Oraichain’s market cap of $118m. Yet, Oraichain’s TVL is $13.2m, which is nearly twice as much as Moonriver’s TVL of $6.78m.

Still, there are L2 networks like Astar Network with nearly double the TVL of Oraichain, yet Oraichain has much higher economic activity.

The network holds a TVL of $13.2m, and 7-day DEX volume of $1.6m, putting its capital efficiency at 0.12. Capital efficiency is a measure of network activity , represented by the transfer of funds –on-chain DEX volume– compared to the chain's TVL.

Simply put, a higher capital efficiency score indicates a higher degree of economic activity (shown by DEX volume).

Another good thing is that ORAI vesting is nearly complete, so when adoption picks up, investors don’t have to be concerned about inflationary unlocks.

Tokens of Interest

A lot of these projects are technically sound and provide solid use cases but they are in the exploratory phase of tokenomics. Projects are having a hard time creating tokenomics that incentivizes use within their ecosystems that, at the same time, embodies the value that the network accrues.

For the near term, there aren’t any tokens I’m interested in. But some of these projects could be interesting to open positions in the mid to long term:  

Mid-term tokens:

1. MOR (Morpheus)
2. ORAI (OraiChain)

Long-term tokens (full tokenomics not released yet or more traction required):

1. ASI (Fetch.ai, Ocean Protocol, and SingularityNet)
2. PROMPT (Wayfinder)
3. TAI (Talus)
4. OLAS (Autonolas)
5. SPEC (Spectral)
6. KAI (Shinkai)

Conclusion

For now, AI agents are gaining more adoption on an industry/developer level, rather than the user level. For instance, Industry leaders and corporations are leaning towards projects like Fetch.ai to integrate their legacy systems with AI, and developers are seeking ways to supercharge on-chain apps with AI/ML pipelines.

Still, more developers need to adopt the tooling of these projects to create value for users and the protocols.

At the moment, projects like Morpheus have already shown that it’s possible to automate simple tasks for on-chain position management with chat interfaces.

Projects building at the L1 level, like Oraichain, are aiming to develop an ecosystem of AI-powered apps like NestStock, an app for AI-powered real estate research, and projects like BlockAI.Care, which provides on-chain pre-diagnosis and health monitoring tools.

In the near future, most transactions on-chain will be submitted by wallets controlled by AI agents. Projects need to ensure their systems are scalable to keep up with high traffic and spam.

All of these AI projects (like the rest of the industry) are experiments. As we’ve seen previously with the rise of Solana – and Arbitrum, Polygon, and other networks– the market is still very early, and the systems still need to be pushed to their limits to determine if they are solid or not.

But if you’re seriously looking to benefit from the AI hype…. Just choose NVDA lol. When Fetch.ai’s ASI is released, I’ll consider picking up some tokens though.