https://github.com/Sentdex/TermGPT

Giving LLMs like GPT-4 the ability to plan and execute terminal commands
https://github.com/Sentdex/TermGPT

Last synced: 8 days ago
JSON representation

Giving LLMs like GPT-4 the ability to plan and execute terminal commands

• Host: GitHub
• URL: https://github.com/Sentdex/TermGPT
• Owner: Sentdex
• License: mit
• Created: 2023-06-03T16:39:05.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-07-20T07:07:48.000Z (over 1 year ago)
• Last Synced: 2024-08-01T19:32:31.444Z (3 months ago)
• Language: Jupyter Notebook
• Size: 21.5 KB
• Stars: 403
• Watchers: 20
• Forks: 95
• Open Issues: 8
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-langchain-zh - TermGPT - 4 等大型语言模型计划和执行终端命令的能力 (其他LLM框架 / 文章)
• awesome-langchain - TermGPT - 4 the ability to plan and execute terminal commands  (Other LLM Frameworks / Videos Playlists)

README

        
# TermGPT

Giving LLMs like GPT-4 the ability to plan and execute terminal commands

Video explanation and usage examples: https://youtu.be/O4EmRi0_CI4

The notebook also breaks down how the script works.

# Usage

Run with `$ python3 TermGPT.py`

You will need to create a `.env` file similar to https://github.com/Sentdex/TermGPT/blob/main/.env.example, or set `OPENAI_API_KEY` manually.

From here, you make your programming/development request. The script will run and query GPT-4 for a series of terminal commands to run to achieve this objective. This is including, but not limited to: reading files, writing code, reading websites, running code, running terminal commands...etc. 

The proposed commands are stored to a list and then presented back to the user again in bold red text, prior to running. After reviewing these commands, you can opt to run them, or not.

# Future work

- I would like to primarily find an open source model that can yield similar performance to GPT-4 in this realm.

- Simplify and add more natural language. Most likely, starting "prompts" will need higher and lower level understanding. Rather than doing something like --r [FILENAME], I would like to have a higher level pass via the LLM to determine if any files should be read, allowing for a pure "natural language" approach, along with likely far more generalization.