Bitcoin x AI: AI Agent Crew

Description

Caution

This is an early experiment, agents do automatic things, here be dragons, run at your own risk.

This repository contains the code for the AI Agent Crew, a collection of agents that can be used to perform assigned tasks using defined tools.

Development

Tech stack:

• Python 3.11 (miniconda recommended for virtual env)
• CrewAI
• CrewAI Tools

To run this locally:

1. Clone this repository
2. Create or activate the virtual environment
   1. Create: conda create -n ai-agent-crew python=3.11
   2. Activate: conda activate ai-agent-crew
3. Install dependencies: pip install -r requirements.txt
4. Run locally: python run_crew.py

Choosing an LLM

CrewAI allows for multiple model configurations, from using OpenAI/Microsoft Azure APIs to running local models, see the documentation for more info on how to configure for your specific needs.

An example env file (.env.example) is provided.

Current testing is being done with Text generation web UI

Command to run Text generation web UI: python server.py --api --verbose

Command to run Text generation web UI on another machine: python server.py --api --verbose --listen

If you already have local models downloaded, you can specify the model directory for Text generation web UI to read from: python server.py --model-dir /path/to/your/model_files --listen --api --verbose

Once running, access the web UI at port 7860 and:

• On the Model tab, select the model to load
• (optional) On the Parameters - Generation tab, set custom parameters for the model
• On the Model tab, load the model and wait until "Successfully loaded" appears

Tip

The IP address and port of the model running on Text generation web UI should be set in .env to use a local model.

Setting up the Wallet

This repository also imports the aibtcdev/agent-tools-ts repository as a submodule.

This provides Typescript functions to interact with a Stacks wallet using Stacks.js.

To clone the repository and sync the submodule, run the following command:

• ssh: git clone --recurse-submodules git@github.com:aibtcdev/ai-agent-crew.git
• https: git clone --recurse-submodules https://github.com/aibtcdev/ai-agent-crew.git

To update the submodule, run the following command: git submodule update --remote --merge

Within the scripts directory is a .env.example file that should be copied to .env and filled out with the wallet information.

Within the scripts/src directory are various scripts that can be run to interact with the wallet. These should be wrapped as a langchain @tool for LLMs to access it.

AI Agent Framework

CrewAI

CrewAI provides an easy-to-use interface for creating and managing agents, tasks, tools, and crews. It is built on top of Langchain, a decentralized, open-source, and privacy-focused AI platform.

• Agents: a team member; an autonomous unit programmed to perform tasks, make decisions, and communication with other agents
• Tasks: a task; individual assignments that agents complete
• Tools: a skill; single-input functions that agents can use to complete tasks (can use any Langchain tool or define custom ones)
• Crew: a collaborative group; a group of agents that work together to complete a set of tasks

Agents

• Agent definitions are in the agents.py file

Tools

• Tool definitions are in the /tools directory