Weave

// Overview

Weave is a Python toolkit from Weights & Biases for building generative AI applications. You add a decorator to the functions you want to track, and Weave logs their inputs, outputs, and the full call tree so you can see what your app actually did at each step.

It is meant for developers working with language models, whether you call hosted APIs like OpenAI, Anthropic, or Google AI Studio, or run open-source models from Hugging Face. As an LLM observability and tracing tool, it gives you a record of every traced call and a place to compare evaluation runs as you move from experiments to production.

Beyond tracing, Weave helps you build repeatable, apples-to-apples evaluations for your use case and keep all the data generated across the LLM workflow organized in one place.

// What it does

• Trace any function by decorating it with @weave.op, building a tree of inputs and outputs
• Works with calls to OpenAI, Anthropic, Google AI Studio, Hugging Face, and other models
• Log and debug language model inputs, outputs, and traces during development
• Build rigorous, apples-to-apples evaluations for your language model use cases
• Organize information across the whole LLM workflow, from experimentation to production
• Free tier available with a Weights & Biases account

// Getting started

You need Python 3.10 or higher and a Weights & Biases account (free tier available). Then install Weave, initialize a project, and decorate the functions you want to trace.

pip install weave

import weave
weave.init("my-project-name")

@weave.op
def my_function():
    # Your tracked code!
    pass

import weave
weave.init("weave-example")

@weave.op
def sum_nine(value_one: int):
    return value_one + 9

@weave.op
def multiply_two(value_two: int):
    return value_two * 2

@weave.op
def main():
    output = sum_nine(3)
    final_output = multiply_two(output)
    return final_output

main()

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Debug a generative AI app by inspecting the recorded inputs, outputs, and traces of each model call
• Compare model or prompt changes with repeatable, apples-to-apples evaluations
• Track an agent's multi-step function calls as a single trace tree
• Keep experiments, evaluations, and production data for an LLM project organized in one place

// How Weave compares

Weave alongside other open-source observability & llmops tools AI/TLDR tracks, ranked by GitHub stars.