BabyCommandAGI

Precautions
- It may unintentionally destroy the environment. Basically, run it in a virtual environment such as Docker.
- It may fail to achieve its purpose and keep looping. (The API usage may increase in such cases, so please use it responsibly.
- Basically, we recommend using Claude 3.7 Sonnet or higher, as it has been verified with Claude 3.7 Sonnet or higher
(If you are using GPT-4 Turbo, it is recommended to use v3 of the older version of BabyCommandAGI)
Objective
BabyCommandAGI is designed to test what happens when you combine CLI and LLM, which are older computer interaction interfaces than GUI. Those who are not familiar with computers may not know, but CLI is an old computer interaction interface. Even now, many computer operations can be done through CLI (common Linux servers mainly use CLI). Imagine LLM and CLI having a conversation. It's exciting to think about what could happen. I hope you will all try it out and find new use cases.
This system is recommended to be run with an API of Claude 3.7 Sonnet or higher.
This Python script system is based on BabyAGI. However, the part that was the thinking part of BabyAGI has been greatly simplified in order to execute commands efficiently. (This may change later)
Use Cases
BabyCommandAGI has the potential to be used in various cases. Please try using it to find use cases.
Below are some known useful use cases.
Automatic Programming
Create an app automatically just by providing feedback
Programming Examples
- Reversi
https://x.com/saten_work/status/1791550524988490053
- Snake Game
https://x.com/saten_work/status/1723509089471492563
Automatic Environment Setup
- Install Flutter in a Linux container environment, create a Flutter app, launch a web server, and make it accessible from outside the container
https://twitter.com/saten_work/status/1667126272072491009
Other
- Get weather forecast
https://x.com/saten_work/status/1791558481432232355
Mechanism
This script works by executing the following continuous loop:
1. pull the next task from the task list. (It starts with one plan task.)
2. determine whether the task is a command task or a planned task
3. if it is a command task:
1. Execute the command.
2. If the Status Code of the command execution result is 0 (success):
Go to 5.
3. Otherwise (failure):
Analyze the history of executions with LLM and create a new task list according to the OBJECTIVE.
4. for plan tasks:
1. plan with LLM based on the plan task, the history of executions and the OBJECTIVE, and create a new task list.
5. If user feedback is generated:
1. plan and create a new task list in LLM based on the OBJECTIVE and the history of executions while being aware of feedback.
Setup
Please follow the steps below:
1.
2. Enter the BabyCommandAGI directory with .
3. Create a file to insert environment variables with .
4. Set ANTHROPIC_API_KEY. (If you use OpenAI models, set OPENAI_API_KEY)
5. (Optional) Set the objective of the task management system to the OBJECTIVE variable.
Execution (Docker)
As a prerequisite, docker and docker-compose must be installed. Docker desktop is the simplest option https://www.docker.com/products/docker-desktop/
Run
Stop
**Note: Even if you exit with Ctrl+C, it will not stop unless you run or . Please be careful.**