Ez OpenAI

My opinion of the openai Python library is best illustrated by the fact that if you ask ChatGPT about it, it will usually hallucinate a more reasonable API. So, I wrote this library, because if I had to manually poll for a tool update again I would instigate the robot uprising myself.

Installation

Run this somewhere:

pip install ez-openai

Usage

Basic usage

Using Ez OpenAI is (hopefully) straightforward, otherwise I've failed at the one thing I've set out to make:

from ez_openai import Assistant

# To use a previously-created assistant:
ass = Assistant.get("asst_someassistantid")

# To create a new one:
ass = Assistant.create(
    name="Weatherperson",
    instructions="You are a helpful weatherperson.",
)

# Create a conversation.
conversation = ass.conversation.create()

# Ask it something.
reply = conversation.ask("How are you today?")

print(reply.text)
"I'm great, thank you! How can I assist you today?"

# You can store the ID for later.
assistant_id = ass.id

# Delete it when you're done.
ass.delete()

Function calling

No more wizardry, just plain Python functions:

from ez_openai import Assistant, openai_function
from enum import Enum

@openai_function(descriptions={
        "city": "The city to get the weather for.",
        "unit": "The temperature unit , either `c` or `f`.",
    })
def get_weather(city: str, unit: Enum("unit", ["c", "f"])):
    """Get the weather for a given city, and in the given unit."""
    # ...do some magic here to get the weather...
    print(f"I'm getting the weather for {city} woooooo")
    return {"temperature": 26, "humidity": "60%"}


ass = Assistant.create(
    name="Weatherperson",
    instructions="You are a helpful weatherperson.",
    functions=[get_weather]
)

# Or, if you already have one, you can fetch it (but still
# need to specify the functions).
ass = Assistant.get("asst_O5ZAsccgOOtgjrcgHhUMloSA", functions=[get_weather])

conversation = ass.conversation.create()


# Similarly, you can store the ID to fetch later:
old_conversation = ass.conversation.get(old_conversation.id)


# The library will handle all the background function calls itself:
conversation.ask("Hi, what's the weather like in Thessaloniki and Athens right now?").text
> I'm getting the weather for Thessaloniki woooooo
> I'm getting the weather for Athens woooooo
> "The weather today in both Thessaloniki and Athens is quite similar, with a
   temperature of 26°C and a humidity level at 60%. Enjoy a pleasant and comfortable
   day!"


# You can give it additional instructions per-run:
conversation.ask(
    "What's the weather like in Thessaloniki?",
    additional_instructions="When asked about the weather, make up something random."
).text
> "Right now in Thessaloniki, it's a pleasant day with clear skies, a gentle breeze,
   and temperatures around 22°C. It's a nice day to explore the city's waterfront!"


# It also supports images:
conversation.ask("What's in this image?", image_url="https://www.someimage.com/").text

# or:
conversation.ask("What's in this image?", image_file="file.jpg").text

Because assistants change (eg if you want to add some more functions), and it's tedious to create new ones every time, there's a helper method that will update an assistant with new functions/instructions:

from ez_openai import Assistant

ass = Assistant.get_and_modify(
    id="asst_someassistantid",
    name="Weatherperson",
    instructions="These are your new instructions.",
    functions=[get_weather, some_new_function]
)

Note: The raw OpenAI message is returned in EZMessage's raw field.

Streaming

If you need to stream tokens, there's a streaming interface:

stream = conversation.ask_stream("Say Hello World!")
for event in stream:
    events.append(event)
message = stream.value

assert "Hello" in events[0].text
assert " World" in events[1].text
assert "!" in events[2].text
assert "Hello World!" in message.text

gg ez