Coinbase AI Agent

This repository demonstrates how to initialize and run an on-chain interaction agent using an open-source language model (Meta-Llama 3.2 1B) combined with the Coinbase Developer Platform (CDP) AgentKit. The agent leverages various toolkits and utilities to interact with on-chain tools, process user input, and provide insightful responses related to blockchain topics.

Usage Guide

1. Download the Repository using

git clone https://github.com/haiderameez/coinbase-ai-agent

2. Install all the dependencies using the following line of code:

pip install -r requirements.txt

3. Environment Variables

• Create a .env file in the root directory and set the following environment variables with your credentials.
• Go to https://docs.cdp.coinbase.com/cdp-apis/docs/welcome and get API keys.
• Download the config.json file and create .env file.
• Store these two in a .env file

CDP_API_KEY_NAME = your_cdp_api_key_name

CDP_API_KEY_PRIVATE_KEY = your_cdp_api_private_key

NETWORK_ID = your_network_id

• Set your own network_id depending on the use case.

4. Setting up Meta-Llama 3.2 1B

• Go to https://huggingface.co/meta-llama/Llama-3.2-1B.

• If permission to use the model is not allowed, fill up the consent form.

• Get access_token from Hugging Face.

• Run the following code,

  huggingface-cli login access_token

5. Understanding the Code Structure

• Model & Tokenizer Setup: The code uses Hugging Face’s transformers library to load the Meta-Llama 3.2 1B model and its tokenizer.

• CDP AgentKit Integration: The CdpAgentkitWrapper is initialized using your API credentials to establish a connection with the CDP ecosystem. The CdpToolkit extracts a set of tools that the agent can use for on-chain interactions.

• Agent Configuration: The agent is created using create_react_agent from the langgraph library, with a specific state_modifier message that configures its on-chain behavior. The configuration includes thread identifiers and checkpoint settings for memory persistence.

• Agent Invocation: The agent receives an input message (e.g., explaining how ERC-721 tokens differ from ERC-20 tokens) and processes it using its integrated tools to generate a response.

6. Running the Agent

• Run the main script using

  python main.py

• This will return an output from the agent.

Please go to these resources to get more information

1. https://docs.cdp.coinbase.com/agentkit
2. https://github.com/coinbase/agentkit
3. https://python.langchain.com/docs/integrations/tools/cdp_agentkit/
4. https://huggingface.co/meta-llama/Llama-3.2-1B