Due to the lack of on-chain transaction interpretation capabilities, we are currently plagued with blind signing, where users sign transactions without understanding their impact. Even with transaction simulation, signers can only verify transactions from the single perspective of balance changes. When facing disguised scams and phishing attempts, relying solely on transaction simulation may mislead signers into believing a transaction is safe. AI-powered DeepTx first simulates transactions and collects context, then conducts analysis and reasoning to explain the potential impact in natural language. This enables users to better understand the transactions they're about to sign, helping them avoid scams and phishing attacks.

DeepTx is AI as a Service, using AI Agents to analyze transaction outcomes in real time and detect fraud and phishing instantly. Before users sign a transaction, it is first sent to DeepTx, which simulates the transaction execution and retrieves relevant data from external tools. Based on this context, DeepTx performs reasoning and explains the transaction's potential impact in natural language, helping users verify if it aligns with their expectations.

The below is our archtechture design.

Etherscan API
Tenderly
Foundry

• Fully AI-Powered
• Analyze potential impacts beyond balance changes
• Real-time analysis without relying on predefined rules, capable of detecting unknown threats
• Provides comprehensive analysis results in natural language

While current mainstream transaction analysis tools focus on transaction simulation, which is static and one-dimensional, DeepTx offers natural language interpretation of broader transaction impacts.

We currently use synchronous logic in our backend code, which prevents us from showing the AI's reasoning process in the UI and requires users to wait for tens of seconds. Additionally, regarding function calling, we haven't achieved complete automation yet and need to improve it.