Ainte

An illegally-made, genetically-engineered, extraterrestrial lifeform which interacts with Ethereum

What is Ainte

Ainte is a command line tool that allows you to interact with the Ethereum blockchain. It is built with Bun and uses an LLM to generate bash commands for you. It is able to interoperate between the EthereumRPC, the Etherscan API and the foundry tooling to get you the information you need.

Vision

Interacting with blockchains is currently a complex process. There are two main methods: blockchain explorers and CLI tools. Explorers are great for basic tasks like calling getter functions or viewing simple transactions, but they're limited for more complex operations. On the other hand, CLI tools like Hardhat and Foundry are more powerful, but they can be cumbersome to use. The vision is to utilize the power of LLMs to interact with the Ethereum blockchain in a more natural and intuitive way.

Hopefully, as LLMs become more powerful, this tool can evolve to a one-stop-shop for the blockchain needs of any developer for both everyday tasks like deploying contracts and decoding transactions, as well as more complex tasks like telemetry data analysis and contract fuzzing.

Status

Currently this is a proof of concept that shows that it is possible to interact with the Ethereum blockchain in a more natural and intuitive way. There is still a lot of work to be done to make this a reality, but the basics are there.

Namely, it is currently possible to:

• Query EthereumRPC for transactions, addresses and more

  • Get me the latest block.
  • Get me the last 10 transactions of address X.
  • How many transactions are there in block X?
  • Get me the balance of address X.

• Query the Etherscan API for contract related information

  • What is the contract name of address X?
  • What is the last transaction of UniswapV3
  • Get me the last 10 events of contract X.
  • What is the contract code of address X?
  • Give me the abi of contract X.
  • Get me the last price of WETH in USD

• Use cast along with the APIs to perform complex tasks

  • What's the current gas price in gwei? (Uses the rpc to query the gas_price and converts it to gwei using cast)
  • How many tokens are minted in WETH? (Uses the rpc api to get the totalSupply of WETH and converts it to decimal using cast)
  • Decode the calldata of transaction X (Uses the etherscan api)

• Answer questions that it already knows the answer to

  • What is the WETH contract address on Ethereum Sepolia?
  • What is the interface of ERC20?
  • What does the swapExactTokensForTokens does in uniswap?

• Perform multistep tasks while keeping context from previous steps

  Example:

  1. Get me the last uniswap v3 transaction.
  2. Decode the calldata of this transaction.
  3. What is this function call exactly?

  Example 2:

  1. Get me the transactions of the latest block
  2. What is the total amount transferred in this block?
  3. Give me this amount in Ether

Apart from these basic functionalities, it also:

• Asks clarifying questions when the task is not well defined.
• Rejects irrelevant questions.
• Corrects itself when the command produced incorrect results.

How it works

This CLI tool creates a blockchain-specialized OpenAI assistant with access to a custom knowledge base including Ethereum RPC, Etherscan API, and Foundry toolset. The assistant provides four types of responses:

1. Direct answers: When the assistant knows the answer to the question, it will return a direct answer.
2. Command answers: When the assistant needs to get something done, it will return the command that the user needs to run.
3. Irrelevant Notices: When the assistant thinks that the question is not relevant to the blockchain domain, it will return an irrelevant notice.
4. Clarifying Questions: When the assistant needs to ask a clarifying question to the user, it will return a message with the question.

The assistant operates through the Ainte CLI tool, which provides a Read-Eval-Print Loop (REPL) for chat interactions. Upon receiving a user message, Ainte propagates the message to the OpenAI assistant and handles the response accordingly. For direct answers and clarifying questions, the assistant will return the answer or question to the user. For command answers, it executes the command directly and returns the output. If a command fails, the assistant automatically retries up to 3 times.

For long outputs the assistant will pretty print the output to the console using jq, less and a markdown viewer.

The tool also supports custom commands, prefixed with /, for quick access to specific functions:

• /exit - Exits the REPL
• /menu - Shows the commands menu
• /retry - Keeps trying to fix the previous commands 3 times in case of error
• /save - Saves the conversation to a file
• /copy - Copies the last output to the clipboard
• /confirm - Enables confirmation prompts before executing commands for our cautious friends
• /noconfirm - Disables confirmation prompts for the brave (the default is being brave)

Setting Up

1. Make sure that you have Node.js, Bun, Foundry, curl and jq and glow installed.
   • For foundry: curl -L https://foundry.paradigm.xyz | bash && foundryup
   • For the rest of the packages:
     • On MacOS with Homebrew: brew install node oven-sh/bun/bun jq glow
     • On Debian/Ubuntu: sudo apt install nodejs jq glow && curl -fsSL https://bun.sh/install | bash
2. Copy .env.example to .env and set your OPENAI_API_KEY.
3. Install the dependencies with bun install
4. You can now run the tool with bun run start.

Keep in mind that the first time you run the tool, it will create an OpenAI assistant and upload the knowledge base to it. Once the assistant is created, put the assistant ID in the .env file, to avoid re-creating the assistant every time.

Docker

1. As described above, copy the current .env.example to .env and set your OPENAI_API_KEY.
2. Build the image with docker build -t ainte .
3. Run the container with docker run --rm -it ainte

Again, keep in mind that the first time you run the tool, it will create an OpenAI assistant. If you want to avoid recreating the assistant every time, put the assistant ID in the .env file and re-build the image.

Communicating with the assistant - Tips and tricks 🕵🏽‍♀️

• The assistant can maintain context from previous steps, allowing you to perform multi-step tasks without repetition.
• To provide additional context, ask relevant questions before stating your main task. For example, instead of directly asking "Get me the calldata of this transaction for this contract," first inquire "What is the interface of this contract?" This approach ensures the assistant has the necessary information (like the ABI) in context before you request the calldata.
• If the assistant makes a mistake that it cannot fix in the first iterations, use the /retry command to make it try again. You can also just rephrase the question, or ask it to "please fix the previous output".

Contributors

We'd love to get contributions from the community! Make sure to check the issues and pull requests to see if there's anything you'd like to work on.

Needless to say, issues and PRs are more than welcome!

Finally, feel free to join our telegram group where all the fun will be happening, or reach out to us on X if you have any questions! 🚀

Acknowledgements

• OpenAI for making accessible the most revolutionary technology of our times.
• Foundry for making the development of this tool easier.
• ArcanEdge for helping us with the LLM stuff since we do blockchains for a living.

From CommonPrefix with ❤️