Towards an Agentic Web3: 2024 Roadmap for Spectral SYNTAX

As ML models become more powerful, the future of onchain engagement will be driven by AI agents. These AI agents, equipped to act autonomously and work with modern cryptography and economic incentives, will advance the Web3 space, bridging the gap between complex onchain activities and user-friendly applications.

At Spectral, we aim to accelerate the Web3 x AI revolution.

Our flagship product, SYNTAX, officially launched in March this year and has captivated the Web3 community by introducing a robust system to generate and deploy smart contracts and fully orchestrated AI agents. Agents like MoonMaker, which carry out onchain operations, and TestMachine, which checks for solidity vulnerabilities, are live on SYNTAX now. We’ve already witnessed users unleash their creativity.

Building on this momentum, we’re doubling down on SYNTAX. This write-up outlines the ambitious roadmap for harnessing the potential of AI agents. From launching our v2 release—a platform where everyone can create and monetize their own agent—to the launch of the Inferchain, a dedicated chain for agent registry and verification, we provide a comprehensive look at the upcoming features designed to expand agent capabilities and their applications.

AI agents in the context of Web3

AI agent is an evolving term, and before we delve further, it may be useful to understand agents conceptually, the components that make up an agentic system, and how these systems operate.

Definition

In the context of Web3, AI agents are autonomous programs driven by models designed to interact with onchain contracts and activities. These agents are equipped with tools to perform complex tasks, bridging the gap between on-chain and off-chain databases, applications, and services.

Core Concepts

1. Instruction Set: A set of goals and guardrails that define the behavior of an agent and dictate the actions an agent can take. For example, "You are an agent that reports prices of cryptocurrencies and executes trades with user approval. Upon receiving the name of a token from a user, you get the latest price and display it to the user. You routinely check the price and ask the user if they want to buy or sell it”.
2. Tools: Each agent is powered by a suite of tools (tools) that enable actions like deploying smart contracts, querying on-chain data, and executing transactions. Agents rely on their instruction set and memory to dynamically select a set of tools depending on a user's request.
3. Wallets: For an AI agent to be truly autonomous, it needs to be able to autonomously sign transactions. With it's own signatory key for a 1-X multisig contract, transaction signing ability can be shared between a user and 1 or many agents.

How Do AI agents Work?

1. An agent is created by setting up the necessary infrastructure, wallets, and tools to perform its instruction set. When a user request is received, the agent uses its tools to gather data required to perceive the environment, processes this data to create a plan, and executes actions based on this plan. For instance, a trading agent might call the Trading View API or get a Uniswap X quote to check for prices and, depending on the instruction set, decide to invoke the Uniswap tool to trade an asset through its agent wallet.
2. Agents can iterate on their actions based on real-time feedback, refining their processes to provide more accurate results. Operating autonomously, these agents manage their own wallets and resources, allowing them to handle funds, pay gas fees, and interact on and off chain without direct user intervention.

Based on experiments at Spectral Labs and in the open-source community, we've concluded that secure and effective agents are possible and upgrade how we interact onchain today. Actions that are impossible for many users today are now accessible through agents that handle complex instruction sets and a versatile set of tools. For more details on the specific workflows of AI agents in SYNTAX, please look out for a technical deep dive, which will be released later this month.

Where are we today?

Since its launch on March 26, 2024, SYNTAX introduced the structure for onchain Agents to thrive—a set of instructions that initializes infrastructure, wallets, and APIs needed to transact onchain. At its core, it is an orchestration system that can route requests between tools and other models, including specialized models designed to generate solidity code.

Using this system, we’ve created some of the first onchain AI agents in Web3, such as SYNTAX MoonMaker and SYNTAX TestMachine. MoonMaker generates code that anyone can use to launch their own memecoin project (from contract deployments to liquidity provisioning to deploying a website) within minutes. 1,000+ memecoins have been launched using MoonMaker and are actively trading on Uniswap. SYNTAX TestMachine helps developers, builders, and users scan for vulnerabilities and identify fixes in smart contracts, helping users remain secure against known exploits.

SYNTAX has captured the mindshare of Web3 users, demonstrating a more performant and seamless way of interacting onchain. Our roadmap aims to build upon this strong initial traction.

Q2: More Agents, More Tools

In the remainder of Q2, we aim to utilize SYNTAX’s code generation and orchestration stack to rapidly increase the number of use cases end-to-end agents can solve and launch at least 10 new versatile agents.

These agents can conduct end-to-end tasks such as exploring onchain data through natural language conversations or simulating trading operations over specific timeframes. By accessing several APIs for onchain and off-chain orchestration, these agents reduce the friction currently experienced when navigating multiple apps, bridges, and protocols, and further enhance the opportunities available to retail users to engage onchain. For example, building a copy-trading agent that mimics onchain wallets based on certain conditions would sound like a daunting task that requires solidity and python skills; with SYNTAX, users can start creating and using such agents within minutes!

We’re also in the early stages of collaborating with other Web3 firms, L1/L2 chains, projects, and communities to develop agents specific to their use cases. These agents can be trained on project-specific documentation, onchain data, and social feeds to create niche workflows which could help in developer relations, business development, liquidity aggregation, and other use cases.

Here are a few examples of upcoming agents on SYNTAX and the variety of use cases they will focus on (exact configurations of agents may be subject to change):

1. Onchain Explorer: Helps users explore onchain data through natural language conversations, accessing the best real time and archived data feeds from DefiLlama, Transpose and Nexandria.
2. Transaction Simulator: Helps users backtest their transaction strategy over a past period of time using scaled data processing through Google BigQuery public onchain datasets.
3. Memecoin Contest Agent: Helps users generate erc20 contracts for memecoins and directly enter them into onchain games.
4. Docs Writer: Trained to write smart contracts along with documentation written at par with best onchain opensource projects. This agent will be tested on upcoming large open source projects, such as Uniswap v4.
5. Filecoin Dev Rel agent: An agent meant to assist storage providers and retail users in supplying storage and deploying content to IPFS network. This Agent uses Filecoin’s Slack channels, partner documentation, and actors on the Filecoin network to answer common questions and interact with the Filecoin Virtual Machine.
6. Social Trend Mapper: Queries popular networks, such as Twitter and Farcaster, to find data and summarize trends related to seminal Web3 events (e.g. a major unstaking event being reported, KOL opinions about a project, etc.). This agent uses data provided by Masa.ai, a decentralized AI data & LLM network.

Agents are more versatile when they have access to a variety of APIs and libraries (a.k.a. tools) that they can use to transact on and off chain. Through Q2, we will be adding several highly used tools across multiple categories such as social networking, market data oracles, trading aggregators, etc. These tools will help us form more partnerships with other projects and communities and strengthen our stack for future product releases in Q3.

1. Price Oracles and API: tools that allow the agent to query market data: 
   1. Chainlink, 
   2. DexScreener
   3. Tradingview
2. JSON-RPC: tools that allow the agent to read blockchain data or send transactions to the network:
   1. Alchemy
   2. QuickNode
   3. Reth Nodes for time sensitive and rpc heavy workflows 
3. DeFi Applications: tools that enable to the agent to interact with top protocols: 
   1. Blur
   2. OpenSea
   3. Uniswap 
   4. 0x
   5. Aave
4. Indexed blockchain data: tools that query onchain data:
   1. DefiLlama
   2. Nexandria
   3. Transpose
5. Decentralized ML Feeds: tools that access inferences from decentralized models:
   1. Bittensor subnets
   2. Challenges on Spectral Nova
   3. Masa.ai feeds
6. Communication and Social: tools that help the agent retrieve and post updates:
   1. Discord
   2. Farcaster
   3. friend.tech
   4. Github
   5. Lens
   6. Slack
   7. Telegram
   8. X / Twitter
7. System Tools: tools which can help process complex workflows:
   1. Foundry
   2. HuggingFace Transformers, TGI, Sentence Transformers
   3. OpenAI

Q3: Build and Monetize Agents with SYNTAX v2

At the start of Q3, we will release SYNTAX v2—which will bring a paradigm shift in how users interact and engage with the onchain agent economy.

1. Create an Agent with AgentBuilder: 
   1. Users can interact with the SYNTAX LLM to create an end-to-end onchain agent. Our agent-building experience is driven by natural language and guides the user to set instructions, knowledge bases, and rules for the agent through a conversational experience.
   2. These agents will have access to all tools, libraries, and APIs we’ve been building this year. Furthermore, users can chat with SYNTAX and set specific triggers and corresponding actions for these agents, making them far more powerful than any alternatives for running onchain operations available today.
   3. Imagine saying, “Hey SYNTAX, build me an ETHBot that trades ETH/USDT anytime ETH is lower than its 200-day moving average,” and voila—you have an agent that does just that, monitoring the markets 24/7!

2. Autonomy through agent wallets: 
   1. Users can build truly autonomous agents using SYNTAX because agents have their own wallets. 
   2. Agent wallets are account-abstracted smart wallets whose private keys are generated in a trusted execution environment, accessible only by the user and the agent; Spectral has no access to these keys. This ensures highly secure operations and allows users to trust their agents to handle funds for running onchain operations.

3. Gas tank: 
   1. Gas costs have always introduced friction in onchain transactions. With agent wallets handling gas costs, this could become a bigger hassle because the user would need to ensure a positive balance of ETH (or any other native EVM token used for gas, such as MATIC) to transact out of the agent wallet. 
   2. To overcome this issue, we’re introducing a new experience—Gas Tank. Gas Tank allows a user to load gas balance using stablecoins and Spectral’s native token, SPEC. Once loaded, this Gas Tank balance will be incrementally deducted to replenish the account abstraction paymaster that helps the Agent execute onchain transactions.
   3. This simple interface will promote gasless transactions, and the user won’t have to worry about swapping assets across chains and sending them back and forth from their personal and agent wallets.

4. Creator-driven onchain agent economy: 
   1. As models become more powerful, we are heading towards an agent-driven future where the next billion users in Web3 will interact with other users and their designated onchain agents. 
   2. We’re taking early steps towards that future by creating agents built for the community by the community! Users can explore the SYNTAX agent Library, which showcases agents built by other community members, and start chatting with any agent instantly. 
   3. Creators of these agents can monetize their agents by receiving a percentage based fee sharing for every transaction their agent makes.
   4. From its inception, Spectral has envisioned the future of Web3 to be truly decentralized. Hence, all agents are open source by nature; any user can fork an existing agent and modify it to create their own agent. This will further incentivize creators to keep updating their agents, making them more performant and purposeful, thus enriching the overall value created in the community.

5. Democratizing governance on the SYNTAX Network: Governance on the SYNTAX network recently went live and is operated through the SPEC token. Creators of agents and users can participate in community governance by suggesting proposals that govern agent operations, integrations with other Web3 projects, fee-sharing mechanisms, etc.

Q4: Trustless Agentic Interactions: The Inferchain

We envision a future where AI agents can verify each other's knowledge and subsequent actions. To that end, we are launching Inferchain in Q4 2024—a dedicated chain for the Web3 community to build onchain agents, engage with them, and verify their activities. While it is still early to describe the intricate engineering behind Inferchain workflows, here is an overview of its differentiating features:

1. Collective Ownership:Truly powerful, well-adopted agents will become the next class of digital assets that communities, investors, and retail users will want to associate with. For this purpose, we’re building cryptographically secure workflows that will allow anyone to stake and acquire collective ownership in any agent of their choice. The agent ownership and its structure will be defined in immutable smart contracts unique for every agent.
2. Agent Node Operators: Agent node operators will be responsible for running the agent instructions (complete onchain and off-chain orchestration) and serving requests incoming from users. These operators will compete among each other for network rewards through a proof-of-stake mechanism verified by the Validators.
3. Agent Network Validators: We’re designing Inferchain to be a proof-of-stake system where Validators ensure that the agent behaves as per its onchain committed instruction set and develops a consensus among other validators in the network.
4. Agent Name Service (ANS): Agent Name Service (ANS) will codify each on-chain agent with its own name descriptor (similar to ENS or DNS domains), providing convenience for users to identify and communicate with an agent.

We’re truly excited for the future of Web3, a world where onchain agents bring exponential value and reduce the barrier to entry for the next billion users. To get a glimpse of this future, sign up on SYNTAX today, and start talking with an agent! Feel free to join our community on Discord or follow us on X to stay in touch with the latest developments on our network.