Skip to main content
At the core of OxVenta’s interaction layer is the AI Agent — a conversational engine that interprets user commands and transforms them into executable DeFi transactions. The agent leverages state-of-the-art language modeling and secure execution to deliver an intuitive DeFi experience.

​
Natural Language Commands (Telegram / Discord)

OxVenta supports user interaction through plain-text input in messaging platforms.

​
Supported Platforms

  • Telegram (live)
  • Discord (planned)
  • X / Twitter DM (planned)

​
Example Commands

Swap 1 ETH to USDT
Create a token called "MyToken" with symbol MYT and 1 million supply
Provide liquidity for MYT/ETH
Check wallet balanceShow my last 5 transactions
The agent will respond with confirmations, status, or prompts for missing info. Future support for voice and multi-language input is planned.

​
How LangChain & Qwen 32B Work Together

The AI agent architecture is powered by two major components:

​
Qwen 32B

  • A large language model (LLM) optimized for command understanding
  • Performs semantic interpretation of user input
  • Handles token extraction, context parsing, and logical inference

​
LangChain

  • Connects the language model with execution tools
  • Manages stateful conversation sessions
  • Provides routing between AI output and blockchain infrastructure
  • Allows function calling (e.g., swap(), deployToken()) via tools
Together, Qwen understands what the user wants, and LangChain figures out how to execute it on-chain.

​
Security & Authentication Flow

Each command is verified and authenticated before any blockchain interaction occurs.

​
Flow

  1. Command received from user (via Telegram/Discord).
  2. Intent parsed and command is validated.
  3. Session wallet or imported key is verified.
  4. Transaction simulation is performed.
  5. Confirmation prompt is sent to user.
  6. Transaction signed and broadcasted.
  7. Success message and transaction hash returned.
No transactions are executed without user approval.

​
Security Features

  • Encrypted session identity (per user)
  • AI-based fraud and anomaly detection
  • Rate-limiting and abuse prevention
  • On-chain audit logging (optional)

​
Prompt Optimization Tips

To improve clarity and performance when interacting with the AI agent:

​
Do

  • Use direct language
    Swap 50 USDT to ETH
  • Be specific with amounts and tokens
    Create token named TravelToken with 1,000,000 supply
  • Specify chains if needed
    Swap 2 ETH to USDC on Base

​
Avoid

  • Vague requests like “Make me rich”
  • Ambiguous token names (“USD” could be multiple assets)
  • Running multiple commands in one sentence
Use /help to view command templates or /history to recall past prompts.