Skip to content

Latest commit

 

History

History
238 lines (196 loc) · 6.27 KB

README.md

File metadata and controls

238 lines (196 loc) · 6.27 KB
Raw

openai-cli

A universal cli for OpenAI, written in BASH.

Features

  • Scalable architecture allows for continuous support of new APIs.
  • Custom API name, version, and all relevant properties.
  • Dry-run mode (without actually initiating API calls) to facilitate debugging of APIs and save costs.

Available APIs:

  • chat/completions (default API)
  • models
  • images/generations
  • embeddings
  • moderations

The default API chat/completions provides:

  • Complete pipelining to interoperate with other applications
  • Allow prompts to be read from command line arguments, file, and stdin
  • Support streaming
  • Support multiple topics
  • Support continuous conversations.
  • Token usage

Installation

  • jq is required.

    • Linux: sudo apt install jq
    • macOS: brew install jq

  • Download script and mark it executable:

    curl -fsSLOJ https://go.janlay.com/openai
chmod +x openai

    You may want to add this file to a directory in $PATH.

    Also install the manual page, e.g.:

    pandoc -s -f markdown -t man README.md > /usr/local/man/man1/openai.1
    Further reading: curl's killer feature -OJ is a killer feature

Now you can try it out!

Tips

Getting started

To begin, type openai -h to access the help manual.

⚠️ If you run openai directly, it may appear to be stuck because it expects prompt content from stdin which is not yet available. To exit, simply press Ctrl+C to interrupt the process.

Why are you so serious?

What happens when the openai command is executed without any parameters? It means that:

  • The default API used will be chat/completions, and the schema version will be v1.
  • The prompt will be read from stdin.
  • The program will wait for input while stdin remains empty.

Quick Examples

The best way to understand how to use openai is to see various usage cases.

  • Debug API data for testing purposes
    openai -n foo bar
  • Say hello to OpenAI
    openai Hello
  • Use another model
    openai +model=gpt-3.5-turbo-0301 Hello
  • Disable streaming, allow for more variation in answer
    openai +stream=false +temperature=1.1 Hello
  • Call another available API
    openai -a models
  • Create a topic named en2fr with initial prompt
    openai @en2fr Translate to French
  • Use existing topic
    openai @en2fr Hello, world!
  • Read prompt from clipboard then send result to another topic
    pbpaste | openai | openai @en2fr

Providing prompt

There are multiple ways to obtain a prompt using openai:

  • Enclose the prompt in single quotes ' or double quotes "
    openai "Please help me translate '你好' into English"
  • Use any argument that does not begin with a minus sign -
    openai Hello, world!
  • Place any arguments after --
    openai -n -- What is the purpose of the -- argument in Linux commands
  • Input from stdin
    echo 'Hello, world!' | openai
  • Specify a file path with -f /path/to/file
    openai -f question.txt
  • Use -f- for input from stdin
    cat question.txt | openai -f-

Choose any one you like :-)

OpenAI key

$OPENAI_API_KEY must be available to use this tool. Prepare your OpenAI key in ~/.profile file by adding this line:

export OPENAI_API_KEY=sk-****

Or you may want to run with a temporary key for one-time use:

OPENAI_API_KEY=sk-**** openai hello

Environment variables can also be set in $HOME/.openai/config.

Testing your API invocations

openai offers a dry-run mode that allows you to test command composition without incurring any costs. Give it a try!

openai -n hello, world!

# This would be same:
openai -n 'hello, world!'
Command and output 
$ openai -n hello, world!
Dry-run mode, no API calls made.

Request URL:
--------------
https://api.openai.com/v1/chat/completions

Authorization:
--------------
Bearer sk-cfw****NYre

Payload:
--------------
{
  "model": "gpt-3.5-turbo",
  "temperature": 0.5,
  "max_tokens": 200,
  "stream": true,
  "messages": [
    {
      "role": "user",
      "content": "hello, world!"
    }
  ]
}
With full pipelining support, you can achieve the same functionality using alternative methods: 
echo 'hello, world!' | openai -n
For BASH gurus

This would be same:

echo 'hello, world!' >hello.txt
openai -n <hello.txt

Even this one:

openai -n <<<'hello, world!'

and this:

openai -n <<(echo 'hello, world!')

It seems you have understood the basic usage. Try to get real answer from OpenAI:

openai hello, world!
Command and output 
 $ openai hello, world!
Hello there! How can I assist you today?

Topics

Topic starts with a @ sign. so openai @translate Hello, world! means calling the specified topic translate.

To create new topic, like translate, with the initial prompt (system role, internally):

openai @translate 'Translate, no other words: Chinese -> English, Non-Chinese -> Chinese'

Then you can use the topic by

openai @translate 'Hello, world!'

You should get answer like 你好，世界！.

Again, to see what happens, use the dry-run mode by adding -n. You will see the payload would be sent:

{
  "model": "gpt-3.5-turbo",
  "temperature": 0.5,
  "max_tokens": 200,
  "stream": true,
  "messages": [
    {
      "role": "system",
      "content": "Translate, no other words: Chinese -> English, Non-Chinese -> Chinese"
    },
    {
      "role": "user",
      "content": "Hello, world!"
    }
  ]
}

Chatting

All use cases above are standalone queries, not converstaions. To chat with OpenAI, use -c. This can also continue existing topic conversation by prepending @topic.

Please note that chat requests will quickly consume tokens, leading to increased costs.

Advanced

To be continued.

Manual

To be continued.

LICENSE

This project uses the MIT license. Please see LICENSE for more information.