zyx 0.2.67

A fun, hyper minimal AI toolkit. Built on the inspirations of marvin-ai.

hammad-dev

the hammad master repository

Libraries

Library	Working Dates	Link	Depreciated
hammadpy	11/2023 -> 01/2024	hammadpy	[x]
hadronic	01/2024 -> 01/2024	hadronic	[x]
suzuka	01/2024 -> 02/2024	suzuka	[x]
yosemite	02/2024 -> 04/2024	yosemite	[x]
zyx	02/2024 -> Current	zyx	[]

zyx - ease of use hyper-fast abstractions

Notebooks	Link
LLM Completions	llm-completions.ipynb
Agents (New)	agents.ipynb
RAG	rag.ipynb
MarvinAI Inspired LLM Functions	llm-functions.ipynb
Multimodal Generations	multimodal.ipynb

Installation

pip install zyx

# New CrewAI Based Agents
pip install 'zyx[agents]'

zyx is a library built with the sole purpose of providing one of the easiest API's and abstractions for working with LLMs, using single functions; to perform tasks such as:

> Click on the function name to view example usage.

• Ease of Use LLM Modules

  • Easy access CLI chatbot (Use either straight from terminal or with the function.)
  • Simple & Modular Chatbot builder to create custom input/output chatbots.
  • Pydantic modules easily accessible @ zyx.BaseModel() & zyx.Field() for structured data.

• Extremely Modular LLM Completions (built on LiteLLM & Instructor)

  • zyx.completion()
  • One function to run the LLM workflow
  • Any LiteLLM supported model is compatible
  • Structured outputs using either Pydantic BaseModels or JSON Dictionaries.
  • Tool calling with Python functions, Pydantic BaseModels & OpenAI format tool/function JSON.
  • Optional tool execution for functions/workflows.
  • Vision model support.

• LLM Agents w/ Multi Agent Pipelines

  • zyx.legacy_agents() - Legacy LLM agents, built on top of instructor.
  • zyx.agents() - New multi agent pipeline, built on CrewAI. (Not available in base package, requires zyx[agents])

• Easy memory & vector search/creation for RAG

  • zyx.qdrant() - Incredibly simple vector store with search & LLM completions.
  • zyx.sql() - Easy SQL data management with search & LLM completions.

• Hyper Quick Multimodal Generations

  • zyx.image() - Image generation connected to Dall-E, FAL-AI, or your local comfyUI or Automatic1111 endpoint.
  • zyx.transcribe() - Speech Transcription OpenAI whisper powered speech transcription; with optional microphone support.
  • zyx.speak() - Text-to-Speech Generations with OpenAI powered TTS, with easy quick-play parameters.

• @Marvin-AI Inspired LLM Functions

  • zyx.function() (comes with two options)
    • Create a mock function response
    • Generate an actual Python method, to return the desired output
  • zyx.classify() - Label based classification.
  • zyx.extract() - Extract content into desired schema.
  • zyx.generate() - Easiest way to create structured data.
  • zyx.chainofthought() - Create a chain of thought response.
  • zyx.code() - Slightly increased code response accuracy.