Author: TrustlessLabs, Medium
AO is built on Arweave's on-chain storage, which enables infinitely scalable decentralized computing, allowing any number of processes to run in parallel at the same time. The decentralized AI Agent is hosted on the chain by AR and run on the chain by AO. AO is an official Arweave project launched by Arweave founder Sam Williams, which supports the introduction of large AI models into smart contracts on the blockchain.
A16Z @a16zcrypto's CTO highly praised AO's programming model as "very elegant" and a truly impressive work. At the same time, Hill, head of research at SevenX Ventures, a well-known Web3 investment institution, excitedly tweeted 20 minutes after the AO conference started: AO may be the ultimate solution for hosting and running decentralized artificial intelligence agents (AI agents).
Storage and computing of on-chain AI
AO's ultimate vision is to achieve seamless integration of AI and blockchain, so that AI models can be hosted on the chain and run directly on the chain, and make autonomous decisions. This will greatly enhance the intelligence level of decentralized applications and bring revolutionary changes to Crypto. Funds are not just passively stored, but can be actively operated, calculated and optimized through Al, which greatly improves the liquidity and utilization efficiency of funds.
In traditional finance, funds are deposited in banks and managed uniformly by banks, which is inefficient and cumbersome. In DeFi, assets can be automatically pledged and circulated, which improves the efficiency of fund utilization. On AO, all funds are managed by users themselves and managed by their own robots or AI. Each fund can have its own strategy, which will greatly improve the intelligence level of financial assets. Because AO has better performance and architecture, users can inject contract code into their wallets and have the agent automatically execute the strategy.
In the past, Arweave, as a storage protocol, could only be seen as a hard drive, and a hard drive alone could not carry a larger narrative and usage scenario. Therefore, Arweave created a CPU that is perfectly compatible with its own hard drive - AO, and integrated it into the narrative of on-chain AI. It can be said that AO saved AR, and the market showed great interest and expectations for AO. Before February 2024, the price of AR had been hovering around $6-10, significantly underperforming BTC, and did not rise with the mainnet upgrade and the start of the bull market. It was not until the release of AO that the price of $AR rose rapidly, rising 4 times in a month. After the news of the $AO token was released, the maximum short-term increase of $AR exceeded 18%. The volatility and wealth-creating effect brought by $AO are becoming more and more obvious.
Challenges and solutions for putting AI models on the blockchain
The core goal of AO is to achieve trustless and collaborative computing services without any practical scale limitations, which provides a new paradigm for applications to combine blockchain. There are many mainstream public chains, but there are many shortcomings for on-chain AI:
Ethereum: Although Ethereum is dominant in smart contracts, its computing and storage space capabilities are limited, making it difficult to handle complex AI models. Compared to Ethereum, which can only use a single shared memory space, AO also allows any number of parallel processes to operate in the computing unit at the same time, and collaborate with other units through open message passing, without relying on centralized memory space. In addition, Ethereum gas is very expensive, and running an AI agent on the entire chain is unacceptable to ordinary retail investors.
Other high-performance blockchains such as Solana, Aptos, and Sui: Despite their high throughput, their main optimization direction is transaction processing speed. They do not provide enough support for complex AI calculations, cannot store AI models on-chain, and their operating environment is not AI-friendly.
Decentralized storage and decentralized computing power such as Filecoin and io.net: Projects such as Filecoin and io.net propose decentralized storage and computing power in an attempt to compete with traditional AWS, but the real value of blockchain lies in consensus and de-trust, rather than direct competition for computing costs or storage costs. Projects such as Filecoin and io.net simply use blockchain technology to implement a storage or computing power market where users can trade, and more often use decentralized technology to allocate computing resources. There is no innovation to achieve censorship-resistant, non-closable or trustless computing methods. Arweave is completely different, focusing on blockchain storage. Blockchain storage is similar to Bitcoin's ledger, with decentralized, tamper-proof and traceable characteristics. In the Arweave ecosystem, there are projects such as Apus that are building AO nodes for GPUs. In the future, these nodes will also run on AO. As a protocol, AO was designed by Sam, the founder of Arweave, in the hope of being more decentralized. Developers around the world can freely develop GPU computing, DPIN and social applications on the AO protocol. It is completely open and does not rely on large VC financing. Instead, it is built in a decentralized way from the beginning. This model makes AO highly decentralized from the beginning.
AO+AR Architecture
AO is a distributed, decentralized, actor-oriented computing system based on Arweave. Actor is the basic unit of a concurrent computing model in computer science. Actor-oriented refers to a method based on the actor model. In this method, each actor can modify its assigned private state, but if the state of other actors is to be modified, it can only be modified indirectly by sending messages. It is suitable for building highly concurrent, distributed, and fault-tolerant systems.
AO consists of three subnets. They include the Messenger Unit (MU), the Scheduling Unit (SU), and the Computing Unit (CU). These three subsystems work together and are responsible for receiving and processing information, scheduling and sorting information, and actual computing operations. Through the ordering of the blockchain, each unit can be used as a horizontally scalable subnet to execute a large number of transactions at the same time, thereby achieving high-performance computing and theoretically providing nearly unlimited computing performance. Its characteristics can be summarized into two:
Separation of computing and storage: AO uses Arweave's decentralized storage solution to ensure the permanence and immutability of data. Arweave is responsible for storing AI models and their related data, ensuring the security and verifiability of data, while AO is responsible for performing complex computing tasks, including reasoning and training of AI models, and achieving verifiability of computing results by storing all interaction logs. This holographic state storage mechanism allows anyone to reproduce the computing process and verify its correctness, which brings new possibilities for decentralized computing.
Unlimited expansion of computing power: AO's design allows any number of processes to run in parallel and coordinates them through a message-passing-based system, which contrasts with the traditional shared memory parallel computing model. This design improves the scalability and efficiency of the system because it avoids the lock contention problem in the traditional model. This is especially important for running large AI models because it can significantly improve computing efficiency and speed.
The key processes in the AO network include: the interactive information in the AO network is passed to the SU after the signature is verified by the MU, and then uploaded to Arweave after being sorted by the SU, and consensus is reached on the order and stored on Arweave. When the user needs to obtain the status, the message will be passed to the SU through the MU. The SU selects the appropriate CU and delivers the message to the CU. The CU completes the status calculation through the data on Arweave and returns the output to the user through the MU.
Sam, the founder of AR, provided a llama3 model on AO in his personal Github repository, which means that large models can be run directly on AO. The AO+AR architecture is tailor-made for on-chain AI, and this design model has now been proven to be very suitable for Al computing. The realization of three key points makes on-chain AI agents truly possible:
Separation of computing and storage: Storing AI models and their training data requires a lot of space. Arweave provides a decentralized and efficient storage solution that ensures data persistence and immutability. Unlike other blockchains, AO can directly perform model training, which is a major breakthrough. AO+AR's SCP (storage consensus paradigm) separates the computing layer from the storage layer, making the storage layer permanently decentralized, while the computing layer maintains the mode of the traditional computing layer. Therefore, AO's computing resources are not essentially different from traditional computing, but a permanent storage layer is added to make traditional computing traceable and decentralized. AO's computing scalability has no type of restrictions, and we can implement multiple computing modes such as EVM and WASM on AO. By storing computing logs and processes on AR, AO's computing power is almost the same as traditional computing. In this way, we can further expand GPU computing and even adapt computing models such as NVIDIA's CUDA, showing its wide scalability.
Actor architecture: On-chain computing is often limited by resources, but AO significantly improves processing efficiency through its parallel computing capabilities. AO itself is designed using the Actor model, which is closely related to AI research. The core idea is that each component of the system can be an independent, autonomous agent, which can be interacted by sending messages when needed. The Actors in this model are very similar to AI Agents. AO can allocate multiple computing units to handle the reasoning tasks of the AI model, thereby overcoming computing bottlenecks.
Security and credibility: To ensure the credibility of AI reasoning results, all interaction logs and calculation processes are recorded on Arweave, and any participant can verify the correctness of the calculation results. AO's security mechanism ensures the credibility of the calculation process and results.
Ecosystem of autonomous AI agents on the chain
Since the release of the testnet, the number of developers and network messages in the AO ecosystem has grown rapidly, and has now exceeded 100 million messages. The AO Ventures incubator has played a key role in promoting the development of the AO ecosystem. By providing technical support and resources, it has attracted a large number of developers and innovative projects, further enriching the AO ecosystem.
Autonomous Finance: Using the AO platform, Autonomous Finance hopes to realize financial agents including: DCA asset management agent, self-balancing index fund agent, autonomous hedge fund agent with customized risk strategy, on-chain prediction agent, high-frequency trading agent, etc. Currently, the DCA investment agent product has been launched, and users can set parameters such as fixed investment asset type, slippage range, liquidity pool, and fixed investment time. Of course, this does not realize the use of AI to formulate investment strategies, but is more at the contract automation level that does not require off-chain triggering.
Apus Network: Trustless GPU solution to realize Trustless AI in AO. Inspired by Sam, Apus Network uses Trustless GPU solution combined with Arweave's decentralized storage and AO's super-parallel computing to achieve verifiable and credible AI inference results, paving the way for building more reliable and transparent AI applications. aos-llama builds AOS images by leveraging the llama2.c inference engine, provides a Lua interface to load the Llama model in Arweave, and provides conversion tools and a comprehensive toolset to efficiently perform AI inference on AO.
In addition to AI projects, there are many good GameFi and DeFi projects on AO:
AO GAMES: AO GAMES is a full-chain game distribution platform built on AO, which will provide Web2-level gaming experience with low entry barriers, extremely fast transaction speeds, and extremely low costs. Importantly, AO GAMES also supports the introduction of AI models to build AI-driven Web3 game projects.
AOX: AOX is the first and only cross-chain bridge project of AO, developed by the everVision team. It provides cross-chain asset services between the AO network and other blockchain networks based on MPC (multi-party secure computing) technology. Currently, AOX is in the Beta stage and only supports the cross-chain function between Arweave and AO networks. At the same time, a cross-chain incentive campaign was launched recently to encourage users to transfer $AR from the Arweave network to AO through AOX. Users earn $TAOX test tokens by completing tasks in order to obtain airdrops of official tokens in the future. It is reported that AOX will successively launch cross-chain services for networks such as Ethereum, BTC and BSC, so that more public chains and more types of assets can seamlessly enter the AO network.
ArSwap: ArSwap is the third DEX to be launched on the AO ecosystem after Permaswap and Bark. It was developed by the Pianity team (Arweave ecosystem music NFT project). Currently, test tokens such as $AR, $AOCRED, $Fire, $Earth, $Water and $Air have been launched. Users can experience transactions and establish LP liquidity pools on ArSwap by receiving faucets. At the same time, the team has recently optimized the product to further increase the speed of transactions.
aoWebWallet: aoWebWallet is the first native AO network wallet, designed to securely manage assets on AO, supporting sending, receiving and tracking AO network assets. Users can log in directly by connecting to the Arconnect wallet (Arweave wallet), or by importing the wallet's .JSON file. Currently, aoWebWallet already supports a variety of AO assets, including $AO-CRED, $Bark, $TRUNK, $0rbit, $Fire and $Earth.
Token Economics
On May 30, 2024, AO announced that it will soon complete the issuance of $AO tokens, and the token will be launched on June 13, Beijing time. In addition, the token related to this project is $AR. Before $AO completes TGE, $AR is still one of the targets of speculation. After the news of $AO tokens was released, $AR's maximum short-term increase exceeded 18%. On the one hand, $AR is currently almost the only target to undertake this event, and on the other hand, it may be related to the rule that holding $AR can obtain $AO tokens. The two rules of bridging assets to AO (currently $AR is basically the only asset that can cross-chain to the AO network) and holding $AR are actually digesting the selling pressure on $AR caused by this event.
Circulation, Token Distribution and Participation Opportunities
According to official news, $AO will be launched with 100% fairness, without pre-mining, pre-sale and priority acquisition, with a total of 21M tokens, which will be halved every four years. The $AO token model adopts the same total of 21 million and halving mechanism as Bitcoin every four years. These design elements not only ensure the scarcity and fairness of the token, but also pay tribute to Bitcoin, demonstrating the recognition of the concept of decentralization and fair distribution.
What is important is the way to obtain tokens: (1) bridge assets to AO (2) hold $AR (3) participate in the construction of the AO ecosystem. However, the official also clearly pointed out that users in the United States and sanctioned regions will not be able to mint $AO through the bridge network.
From the release method, it can be seen that the official hopes to ensure that the token can be launched 100% fairly. It is worth noting that cross-chain to AO is one of the ways to mint tokens. Currently, AOX ( aox.xyz ) is the only cross-chain bridge in the AO ecosystem.
Token usage scenarios
According to the answer of outprog, one of the proposers of AO, in X Space, AO tokens and AR tokens will have their respective functions. AR tokens will mainly focus on the storage function and consensus maintenance of Arweave, while AO tokens will focus on solving the communication problems between computing and applications. That is, AO and AR will maintain the computing and storage functions of the network respectively.
