Easier OpenAI

Easier OpenAI wraps the official OpenAI Python SDK so you can drive modern assistants, manage tool selection, search files, and work with speech from one helper package -- all with minimal boilerplate.

Features

• High level Assistant wrapper with conversation memory, tool toggles, and streaming helpers.
• Temporary vector store ingestion to ground answers in local notes or documents.
• Text to speech and speech to text bridges designed for quick experiments or internal tooling.
• Built-in helper for defining and executing OpenAI function tools without leaving Python.
• Lazy module loading so import easier_openai stays fast even as optional helpers expand.
• Type hints and comprehensive inline docstrings across the project for easier discovery.

Installation

pip install easier-openai

Optional extras:

pip install "easier-openai[function_tools]"   # decorator helpers
pip install "easier-openai[speech_models]"    # whisper speech recognition models

Set the OPENAI_API_KEY environment variable or pass api_key directly when instantiating Assistant.

Quick Start

from easier_openai import Assistant

assistant = Assistant(model="gpt-4o-mini", system_prompt="You are concise.")
response_text = assistant.chat("Summarize Rayleigh scattering in one sentence.")
print(response_text)

sped-up video. actual time ~1 min:

quickstart.mp4

Tool Calling Made Simple

Use Assistant.openai_function to convert regular functions into structured tool definitions and hand them to chat:

from easier_openai import Assistant

assistant = Assistant()

@assistant.openai_function
def look_up_fact(topic: str) -> dict:
    """Return a knowledge base lookup result for the given topic."""
    return {"topic": topic}

assistant.chat(
    "Tell me about the ozone layer using the fact tool.",
    custom_tools=[look_up_fact],
)

Stream Responses with Tools

stream = assistant.chat(
    "Summarise launch blockers for the robotics demo",
    custom_tools=[look_up_fact],
    text_stream=True,
)
for delta in stream:
    if delta == "done":
        break
    print(delta, end="", flush=True)

Ground Responses With Your Files

notes = ["notes/overview.md", "notes/data-sheet.pdf"]
reply = assistant.chat(
    "Highlight key risks from the attached docs",
    file_search=notes,
    tools_required="auto",
)
print(reply)

Speech I/O

Generate audio output directly from assistant responses:

assistant.full_text_to_speech(
    "Ship a status update that sounds upbeat",
    model="gpt-4o-mini-tts",
    voice="alloy",
    play=True,
)

full_text_to_speech accepts the same keyword arguments as chat, so you can pass custom_tools, file_search, or web_search before the reply is spoken.

Or capture short dictated prompts without leaving the terminal:

transcript = assistant.speech_to_text(mode="vad", model="base.en")
print(transcript)

Image Utilities

Openai_Images extends the assistant with helpers that accept URLs, file paths, or base64 payloads and normalise them for the Images API:

from easier_openai import Openai_Images

image_client = Openai_Images("samples/promenade.jpg")
# Generated metadata is stored on image_client.image for re-use in calls.

Configuration Reference

• model: Default model used for chat, tool calls, and reasoning workflows. Choosing a realtime model (for example gpt-4o-realtime-preview) automatically routes Assistant.chat through the Realtime API for text-only prompts; install openai[realtime] to satisfy its websocket dependency. Tool execution and the stream flag are ignored while a realtime model is active.
• system_prompt: Injected once per conversation to shape assistant behaviour.
• reasoning_effort and summary_length: Fine tune reasoning models via the official API semantics.
• temperature: Pass through value mapped to OpenAI responses for deterministic vs creative answers.
• function_call_list: Pre-register decorated tool callables that should accompany every chat request.
• default_conversation: Set to False if you prefer to supply conversation IDs manually.
• mass_update(**kwargs): Bulk update configuration attributes using keyword arguments validated by type hints.

assistant.mass_update(model="gpt-4o-mini", temperature=0.2)
assistant.mass_update(reasoning_effort="high", summary_length="concise")

Developer Notes

• Every public function and class ships with contextual docstrings to make the codebase self-documenting.
• The repository includes unit tests under tests/ that exercise tool-calling flows; run them with pytest.
• Generated artifacts in build/ mirror the source package and inherit the same documentation updates.
• Issues and pull requests are welcome; please run checks locally before submitting changes.

License

Licensed under the Apache License 2.0.