In my never-ending quest to minimize context switching when writing code, I wrote a bash script that allows you to call the OpenAI API chat endpoint directly from the command line with questions that would otherwise take you to Google, followed by Stack Overflow (think "overflow python format timestamp mm-dd-yyyy" or "overflow git edit commit message").

I have only tested this in Ubuntu 22.04, but if you park said script (which I called overflow in keeping with the original goal) in /usr/bin and then run chmod 755 overflow, you can do things like this:


or this:


or this:


and of course, our first example:


At least in these simple examples, all of the generated code runs as-is. Of course, the possibility of hallucination is always present, and all outputs should be used with discretion.


The script itself is very simple:

input="${@}?"

if [ ${OPENAI_API_KEY+x} ]; then
  echo
  result=$(curl -s https://api.openai.com/v1/chat/completions \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $OPENAI_API_KEY" \
    -d '{
      "model": "gpt-3.5-turbo",
      "messages": [
      {
        "role": "system",
        "content": "The user is a programmer interfacing with this LLM via the command line. They want to know more information about the provided input. The input may be a question, or simply a sentence fragment. If it is a fragment, they probably want to know how to implement the stated functionality. Or, they may be trying to avoid the stated issue from happening. Please give as concise a response to the input as possible in the style of a response on Stack Overflow - output is limited to 500 tokens. Do not add any more words beyond what is relevant to the response." 
      }, { 
        "role": "user",
        "content": "'"$input"'"
      } 
      ],   
      "temperature": 0.7,
      "max_tokens": 500
    }' | jq -r '.choices[0].message.content')
  echo -e "$result"
else 
  echo "need to set envvar 'OPENAI_API_KEY'!"
fi

It requires you to set an envvar, OPENAI_API_KEY, to a valid OpenAI API key. It also has one dependency, jq, for parsing the JSON output of the API call and extracting only the response of the model.

The prompt is simply a combination of the user's input with a question mark appended to the end, along with the below. Whether this prompt actually confers any improvement to the output requires more testing.

The user is a programmer interfacing with this LLM via the command line. They want to know more information about the provided input. The input may be a question, or simply a sentence fragment. If it is a fragment, they probably want to know how to implement the stated functionality. Or, they may be trying to avoid the stated issue from happening. Please give as concise a response to the input as possible in the style of a response on Stack Overflow - output is limited to 500 tokens. Do not say anything beyond what is relevant to the response.

Despite the emphatic wording, however, the prompt does not stop the model from... deviating from its instructions 😆