[!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.
Tech stack:
To run this locally:
conda create -n ai-agent-crew python=3.11
conda activate ai-agent-crew
pip install -r requirements.txt
python run_crew.py
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:
Model
tab, select the model to loadParameters - Generation
tab, set custom parameters for the modelModel
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.
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:
git clone --recurse-submodules git@github.com:aibtcdev/ai-agent-crew.git
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.
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.py
file/tools
directory