brokit 0.1.1

brokit, think it like a playing lego

brokit

Inspired by big bro DSPy, brokit is a minimal Python toolkit of composable, LEGO-like primitives for working with language models. Build what you need, skip the bloat.

What's This About?

A lightweight library for working with LMs across any use case. Just the essential building blocks, nothing more.

Core Concepts

Coming from DSPy? You already know what's up:

• Prompt = dspy.Signature — Define your input/output structure
• Predictor = dspy.Predict — Execute prompts with your LM
• LM = dspy.LM — Language model interface
• Shot — Few-shot examples made simple

Design Philosophy

Plug and Play

Everything's a base class. Compose, extend, swap out whatever you want. The LM module? Bring your own.

Pure Python

Zero required dependencies. Want to use requests, httpx, or boto3? Go for it. Check the notebooks for integration examples.

Features

• Text and image support (more formats coming)
• Few-shot learning with Shot
• Build custom LM implementations
• Structured prompts with type hints

Installation

pip install brokit

Quick Start

import brokit as bk

# Define your prompt structure
class QA(bk.Prompt):
    """Answer questions"""
    question: str = bk.InputField()
    answer: str = bk.OutputField()

# Create your LM and predictor
lm = YourLM(model_name="your-model")
qa = bk.Predictor(prompt=QA, lm=lm)

# Get results
response = qa(question="What is brokit?")

Cookbook

Check out the cookbook/ directory for hands-on examples:

LM Integrations:

• lm/ollama.ipynb - Ollama integration with vision support
• lm/bedrock.ipynb - AWS Bedrock integration

Predictor Patterns:

• predictor/zero_shot.ipynb - Basic prompting and chain-of-thought
• predictor/few_shots.ipynb - Few-shot learning with examples

What's Next?

Check out ROADMAP.md for what's coming and VERSIONS.md for release notes.