bash-gpt

README

BashGPT

A GPT assistant for the command line.
Tuned to assist with developer tasks like finding files, installing packages, and git.

Usage

• Install bashgpt:

  go install github.com/Ztkent/bash-gpt/cmd/bashgpt@latest
  

• Set your API key as an environment variable:

  export OPENAI_API_KEY=<your key>
  export ANYSCALE_API_KEY=<your key>
  

• Run the assistant:

  # Ask the assistant a question
  bashgpt [your message]
  
  # Start a conversation with the assistant
  bashgpt -c
  bashgpt -llm=openai -c -max-messages=250 -max-tokens=1000 -t=0.5
  

Examples

bashgpt install Python 3.9 on Ubuntu
- sudo apt update && sudo apt install python3.9

bashgpt given a text file, wrap each line in quotes. format it for display
- sed 's/.*/"&"/' file.txt

bashgpt given a text file, wrap each line in quotes. format it for display with python
- with open('file.txt', 'r') as f:
  lines = f.readlines()
  lines = ['"{}"'.format(line.strip()) for line in lines]
  print('\n'.join(lines))

bashgpt update git email and username
- git config --global user.email "youremail@example.com"
  git config --global user.name "Your Name"

bashgpt git re-edit a specific commit
- git commit --amend

Options

• There are two options for the API provider:
  • OpenAI (https://platform.openai.com/docs/overview)
  • Anyscale (https://www.anyscale.com/endpoints)

Flags:
  -h:                        Show this message
  -c:                        Start a conversation with BashGPT
  -llm [openai, anyscale]:   Set the LLM Provider
  -m [string]:               Set the model to use for the LLM response
  -max-messages [int]:       Set the maximum conversation context length
  -max-tokens [int]:         Set the maximum number of tokens to generate per response
  -t [0.0-1.0]:              Set the temperature for the LLM response
  -d:                        Show debug logging

Model Options:
  - OpenAI:
    - gpt-3.5-turbo, aka: turbo
  - Anyscale:
    - mistralai/Mistral-7B-Instruct-v0.1, aka: m7b
    - mistralai/Mixtral-8x7B-Instruct-v0.1, aka: m8x7b
    - meta-llama/Llama-2-7b-chat-hf, aka: l7b
    - meta-llama/Llama-2-13b-chat-hf, aka: l13b
    - meta-llama/Llama-2-70b-chat-hf, aka: l70b
    - codellama/CodeLlama-34b-Instruct-hf, aka: cl34b
    - codellama/CodeLlama-70b-Instruct-hf, aka: cl70b

API Provider

By default the assistant will use the Anyscale API. To use OpenAI, run the assistant with a flag.

bashgpt -llm=openai
bashgpt -llm=anyscale 

Model

Depending on the LLM Provider selected, different models are available.
By default the Anyscale API uses Mistral-8x7b, and OpenAI uses gpt-3.5-turbo.

bashgpt -m=m8x7b

Conversation

The assistant can be used in conversation mode.
This allows the assistant to remember previous messages and use them to generate more in-depth responses.

bashgpt -c

Conversation Context

Larger conversations require more tokens, by default the conversation context is limited to 100 messages.

bashgpt -max-messages=250

Token Limit

Tokens cost money, by default the assistant will generate as many tokens as it needs for the converation.

bashgpt -max-tokens=10000

Temperature

The temperature of the LLM response is a measure of randomness. Adjust this value to taste. Temperature is a float between 0 and 1. By default the temperature is 0.2

bashgpt -t=0.5