Building a Hyperliquid AI Agent Trader From Scratch

Introduction to HyperLiquid and Its Expanding Offerings

• HyperLiquid is a platform that has expanded its offerings beyond cryptocurrency trading, now including more interesting assets such as OpenAI and the S&P 500, as well as Brent oil, which is a perp market deployed independently by YEY 42s.
• To set up an AI agent on HyperLiquid, it is necessary to create an account and connect it using options such as MetaMask, WalletConnect, or an email account, with MetaMask being a recommended choice 2m6s.
• After connecting the wallet, the next step is to deposit funds into the account, which requires using the Arbitrum network or Hyper EVM Arbitrum, and having assets such as USDC and ether 4m30s.
• To deposit funds, one can use MetaMask to buy USDC on the Arbitrum network, and then swap some of the USDC for ether to cover gas costs, with the process involving selecting the Arbitrum network, choosing USDC, and using a payment method such as Revolut Pay 6m15s.

Setting Up a Wallet and Connecting to HyperLiquid

• Once the USDC and ether are in the MetaMask wallet, the funds can be deposited into the HyperLiquid account, allowing the AI agent to start trading 10m30s.
• The entire process involves creating a wallet, connecting it to HyperLiquid, depositing funds, and setting up the AI agent, with the goal of automating trades on the platform 12m0s.
• To set up a Hyperliquid AI agent trader, it is necessary to create a wallet and generate a private key, which should be saved in a secure location, such as a .env file, for later use 10s.
• The .env file should contain the API wallet name, wallet address, and private key, and this setup is essential for the trader to function properly 1m30s.
• Before trading can begin, some cash needs to be deposited into the wallet, as no trades can be made without it, and the documentation for Hyperliquid can be found and saved for reference 2m40s.
• Once the wallet is set up and cash is deposited, the account will show the wallet names, addresses, and available cash, indicating that it is ready for trading 4m10s.

Depositing Funds and Preparing for Trading

• To start trading, it is recommended to have an agentic setup that can leverage models like Opus, Sonnet, or Codex, and to test the API connection by reading environment variables and checking if the API is live 6m20s.
• The trader can be set up to use either spot or perp trades, and to switch between them, the HIP 3 DEX abstraction needs to be disabled in the settings, allowing for the activation of the perps to spot button 8m50s.
• With the setup complete and the API connection tested, the trader is ready to start making trades on Hyperliquid, with both spot and perp options available 10m40s.

Configuring the AI Agent for Trading

• The setup for the agent is complete, and it's now possible to start setting up interesting trades, with the option to switch to perpetuals and be ready to go, allowing for the creation of a trader with a specific profile and skills 10s.
• A meme role for the trader was created, called the Wall Street Bet Moderator, with an RPG profile, core stats such as risk tolerance and conviction, and combat skills, which will reflect the trades made and add a fun element to the process 1m30s.
• The trader profile has an ultimate mode, 999 X G Gen mode, and will affect the trades made, with the option to create a serious profile for those who want to use this for a side hustle or other purposes 2m6s.

Designing AI Agent Skills and Trading Strategies

• Two skills were created for the agent: find trades and research idea, with the find trades skill being an ID generation funnel for hyperliquid, looking for intraday trades and high-frequency trading opportunities 3m30s.
• The research trade skill will take the ID generated by the find trades skill and research it using various tools, including browser use and sub-agents, to provide more information on potential trades 4m40s.
• The pipeline for finding trades was created in Claude Code, with a straightforward process that starts by reading the Claude MD and running the find trades skill, which can be replicated every time to generate new trade ideas 6m0s.
• The find trades skill uses a Kanban board and runs a bash command to ground the research in the current time, spinning up the research and firing off six parallel sub-agents to look for an ID idea for trading 7m30s.
• The system generates five trade ideas, including long VLD, long pit, Nvidia post print directional reaction, long ADI, and short PER, and then rates each idea based on how well it reflects the trading profile, with the Nvidia post print pre-spec reaction getting a 9.5 out of 10 rating 10s.

Refining Trade Ideas and Filtering Opportunities

• The system excludes some trade ideas, specifically IDs two and five, in order to focus on more complex and interesting trades, and then generates new ideas, including the Nvidia iron triangle, a semi-desperation rest strategy, and the cross catalyst pincer 2m6s.
• The system researches the new trade ideas, including long Nvidia, short AMD, short MU, and long VLD, short BTC, and then triggers the research ID skill to find the best trade with the most expected value and potential 4m30s.
• The system uses various tools, including the browser and the Polymarket API, to conduct research on the trade ideas, including checking Reddit and Polymarket for information on Nvidia, and then gathers context data from various sources 6m20s.

Conducting Research and Finalizing Trade Decisions

• The system continues its research and gathers more information, including looking at Nvidia predictions in tech, and then finalizes the trade brief, which includes allocating 33% of the book and using 20x leverage 10m10s.
• The system is set to execute the final trade, with the user instructing Claude Code to fire the trade, and the system is designed to work autonomously, allowing the user to come back to a completed trade at the end of the cycle 12m30s.

Executing and Testing the Trade

• The trade was originally designed to trigger later, but it was executed immediately for the sake of the video, and a 20x short leverage on Nvidia was planned, but only 10x leverage was used for the test 10s.
• The agent was able to execute the trade programmatically, and the result was a 10x short on Nvidia, which was then closed immediately after the test 42s.
• The platform used for the trade is Hyperliquid, which allows users to learn about financial markets and set up AI agents, and it is considered a great place to learn about these topics 2m6s.

Reflections and Educational Goals

• The user had created a meme profile, but the main goal is to have fun and learn about financial markets, and Hyperliquid is considered a better platform for learning than Polymarket 2m6s.
• The user plans to continue learning about Hyperliquid and setting up AI agents, and possibly comparing the performance of different models, such as Codex versus Claude code 4m30s.

Promotion and Community Engagement

• The video is part of a treasure hunt, and viewers can find more information about it on the user's X account, where they can follow the clues and solve the puzzle to find a reward 6m40s.
• The user invites viewers to check out the treasure hunt, leave comments, like the video, or subscribe to see more content about Hyperliquid and AI agents in the future 8m10s.