askgpt

askgpt is a GPT-based chatbot command line tool that sends a prompt to multiple LLMs (e.g. OpenAI's gpt-4 and Bard's text-bison-001) and displays all the responses. Conversational memory allows follow-up questions to previous questions by sending context as part of the prompt.

Example session

This demonstrates the ability to ask follow-up questions and query multiple LLMs simultaneously.

$ askgpt --gpt4 "who won the superbowl in 2006?"
A(gpt-4):The Super Bowl in 2006 (Super Bowl XL) was won by the Pittsburgh Steelers.

# you can ask follow up questions that recall earlier parts of the conversation
$ askgpt --gpt4 "and in 2008?"
A(gpt-4): The Super Bowl in 2008 (Super Bowl XLII) was won by the New York Giants.

# and retain that conversation across different models
$ askgpt "and in 2009?"
A(text-davinci-003): The Super Bowl in 2009 (Super Bowl XLIII) was won by the Pittsburgh Steelers.
A(gpt-4): The Super Bowl in 2009 (Super Bowl XLIII) was won by the Pittsburgh Steelers.
A(gpt-3.5-turbo): The Super Bowl in 2009 (Super Bowl XLIII) was won by the Pittsburgh Steelers.
A(bard): The Super Bowl in 2009 (Super Bowl XLIII) was won by the Pittsburgh Steelers.

Usage

Usage: askgpt [OPTIONS] PROMPT
    OPTIONS:
        --skip-history  Skip reading and writing to the history.
        --gpt4          Shortcut to set LLM_MODELS=gpt-4
        --bard          Shortcut to set LLM_MODELS=bard
        --info          Set log verbosity to info level
    PROMPT              A string prompt to send to the GPT models.

Environment variables:
  PROMPT_PREFIX       A prefix to add to the prompt read from STDIN.
  OPENAI_API_KEY      API key for OpenAI
  BARDAI_API_KEY      API key for Bard AI
  LLM_MODELS          Comma-separated list of LLM models
  MAX_TOKENS          Maximum number of tokens for a prompt
  HISTORY_LINE_COUNT  How many lines of history context should be considered to add to the prompt

Examples:
  askgpt "Generate go code to iterate over a list"
  askgpt "Refactor that generated code to be in a function named Scan()"
  cat main.go | PROMPT_PREFIX="Generate a code review: " askgpt
  askgpt --skip-history --gpt4 --info "Generate go code to iterate over a list"

The program uses a history to provide prior question context to the current prompt, enabling a short-term memory across sessions that feels like a conversational chat.

Available Models

askgpt supports the following OpenAI GPT models:

openai.GPT3Dot5Turbo
openai.GPT4
text-davinci-003

It also supports the bard model from Google.

Output

The program will output each response from the models along with their respective model names. The program will also keep a history of all questions asked and answered in ~/.askgpt_history to provide context going forward for succcessive questions.

Installation

Clone the repository:

git clone https://github.com/derwiki/askgpt

Build the executable file:
```
go build main.go
```

Run the program:

./askgpt "your prompt here"

Or, you can pipe a prompt to STDIN:

echo "your prompt here" | ./askgpt

And optionally add a prompt prefix:

cat main.go | PROMPT_PREFIX="generate a README.md for this program.\n\nProgram source: ""

Send to a subset of models:

LLM_MODELS=gpt-4 ./askgpt "how do I read an environment variable?"
LLM_MODELS=gpt-4,gpt-3.5-turbo ./askgpt "how do I read an environment variable?"

License

This project is licensed under the MIT License - see the LICENSE file for details.

derwiki / askgpt