LLM

• It is the front-facing component that interacts directly with user input

• Leverages OpenAI, deepseek LLM to handle general-purpose communication and initial decoding of user queries

• Converts unstructured user input (e.g., "Maximize yield on my USDC") into structured intents

Key Features

• Natural Language Processing

  • We try to understand what the user wants

  • Example: Swap ETH for USDC from Ethereum To Solana

    action: SWAP,
    token_in: ETH,
    souce_chain: Ethereum
    token_out: USDC,
    destination_chain: Solana

• Contextual Awareness

  • Captures session-specific context

  • e.g., a prior mention of “Solana” sets the default chain for subsequent action

  • Avoids cross-session data retention to ensure privacy

• Ephemeral Processing

  • Operates entirely in volatile memory—no user input or intermediate data is stored