🇧🇪 ETHGlobal Brussels 2024 - Near AI Assistant

This project aims to provide a one-stop AI chatbot to discover the Near ecosystem 🚀 It was built during the ETHGlobal Brussels 2024 hackathon and won the Near prizes in the AI track.

It features an AI agent using an open-source model running on the Aleph.im decentralized cloud, and a basic frontend to simplify interactions with it.

The agent has access to several functions to serve requests about Near as best as possible:

• Access to realtime search results (Google), which is used to answer questions related to general Near architecture
• Ability to send testnet NEAR tokens to an address when a user requests it
• Capable of minting and sending a special ETHGlobal Brussels NFT on Near testnet when a user requests it
• Able to simplify the understanding of what happened in a testnet transaction with a given hash and sender

❓ It's always better with examples, so here are some questions that this AI is able to handle:

• What is Near?
• What are the different transaction actions on Near?
• Can I please have an ETHGlobal Brussels NFT sent to me at rezarah.testnet? Thanks
• I want to start using Near, can you send me some tokens on my testnet address random.testnet?
• I don't understand what this transaction is doing, can you help me? The transaction hash is hash and it was sent by someone.testnet.

⚙️ Installation

AI

Setup your Python environment with these commands:

cd ai/
python -m venv venv
python -m pip install poetry
poetry install

Then you can create a .env file and fill it with values inspired from the .env.example file.

Front

Setup your frontend with these commands:

cd front/
yarn

🚀 Getting started

Now you're only two command away from using your Near AI assistant 🔥

Start the AI

python src/main.py

and launch the frontend

yarn dev

📈 Current caveats and future improvements

As a hackathon project built in solo, this is obviously not finished and can be improved in many ways. This section will focus on the AI / features side, but the frontend and the way of interacting with this AI can of course also be enhanced.

• Better / longer context and retry
  The model currently used is Nous Hermes 2 Pro - Llama 3 8B, which is really good for handling JSON and function calls in particular (that's why I chose it for this hackathon). While it produces great results, the 8192 context length can be limiting. Having models with higher context length could allow for chat history persistence, more recursive / retry loops etc (but of course such models like Phi 3 with its 128k context length have worst results with JSON, XML, function calls etc).

• Fully autonomous and on-chain agent
  Having the AI inference done in a decentralize way is already a big step forward a fully decentralized AI agent. As it's already connected to the blockchain, able to control a wallet and make transactions, a future where we have an agent that provide services to people, receive tokens for it, and use them to pay for its own computing on a decentralized cloud like Aleph is really near (no pun intended 🤪)

• Better integration with the Near ecosystem
  This is the main obvious point, but currently the action done by the AI on-chain are pretty basic. Developing other complex tools available for the AI would really make it an interesting project to become the one-stop AI chatbot for Near 🚀

➕ Miscellaneous

Other things I have used / fixed related to Near during this hackathon

• Made a fork of py-near, based on another fork that was solving some dependencies issues: https://github.com/RezaRahemtola/py-near
  This allowed me to use the latest version of py-near without the dependency issues from the base project.

🧠 I found this py-near package really cool (great typing) and useful, but to be perfect a few things are missing imho:

• Add it to the near organization and maintain it if possible, currently there is an outdated API package there linking to py-near. Having someone maintaining it would avoid this "fork of a fork of a replacement of near-api-py" that I had to made
• Enhance the error handling, some error messages are really unclear
  • For example, a wrong contract_id param in the account.function_call method gave me... a Rust error (from the .so of the core primitives) about transaction.get_hash() returning a wrong value
• Could be a great thing to add to the awesome-near repo in a Libraries / Frameworks section

• Some small PRs for outdated documentations or typos:
  • near/docs#2139
  • pvolnov/py-near#22

Made with ❤️ by

_{Reza Rahemtola}