AWS re:Invent 2025 - Autonomous Web3 agents on AWS (DAT457)

Summary of AWS re:Invent 2025 - Autonomous Web3 Agents on AWS (DAT457)

Overview

This presentation showcased the development of a web3 assistant agent built using AWS Bedrock and Agent Core primitives. The agent is designed to provide valuable context and insights for web3 enthusiasts by:

1. Accessing and processing blockchain data and market information
2. Maintaining stateful memory to remember user preferences and conversation history
3. Integrating with specialized agents to perform tasks like signing and broadcasting transactions

Key Features and Capabilities

Blockchain and Market Data Integration

• The agent can retrieve real-time data from sources like CoinMarketCap using a headless browser integration
• This allows the agent to gather unstructured data on prices, sentiment, news, and other market insights

Stateful Memory and Personalization

• The agent uses Bedrock Agent Core's memory capabilities to store user preferences, conversation history, and other contextual information
• This allows the agent to personalize its responses and maintain continuity across interactions

Agent-to-Agent (A2A) Integration

• The agent can invoke specialized agents to perform blockchain-specific tasks like signing transactions using AWS KMS
• This allows the agent to leverage the capabilities of other agents without needing to implement the low-level blockchain integrations itself

Deployment and Infrastructure

• The agent is packaged and deployed as a Docker container, allowing for easy CI/CD integration
• AWS Bedrock Agent Core Runtime is used to manage the agent's execution, providing features like session isolation and versioning

Technical Details

• The agent is built using the Strands Agent SDK, which provides an asynchronous, streaming interface
• Bedrock Agent Core's memory capabilities are leveraged to store user preferences and conversation history
• A custom memory hooks provider is used to integrate the agent's memory with the Bedrock Agent Core runtime
• The agent utilizes a headless browser integration (using Playwright) to gather unstructured data from web sources
• For blockchain interactions, the agent invokes a specialized "crypto AI agent" that handles tasks like signing transactions using AWS KMS

Business Impact and Use Cases

• The web3 assistant agent can provide valuable insights and context for web3 enthusiasts, helping them stay informed on market trends, news, and their personal portfolio
• By integrating with specialized blockchain agents, the assistant can also facilitate tasks like checking balances, sending transactions, and managing wallets, all within a conversational interface
• The agent's stateful memory and personalization capabilities allow it to build a deeper understanding of the user's preferences and needs over time, enhancing the overall user experience

Example Interaction

• The user asks the agent to "Provide some market data for my crypto"
• The agent uses its browser integration to gather data from CoinMarketCap, including prices, trading volume, and sentiment analysis for the user's portfolio (Bitcoin, Ethereum, Solana)
• The agent then summarizes the key insights and provides the user with an overview of the current market conditions

Conclusion

The web3 assistant agent showcased in this presentation demonstrates the power of combining Bedrock Agent Core's primitives with specialized integrations to create a valuable tool for web3 enthusiasts. By seamlessly accessing blockchain data, maintaining user context, and leveraging other agents' capabilities, this agent can provide a personalized and informative experience for managing one's web3 activities.