> For the complete documentation index, see [llms.txt](https://docs.three.dev/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.three.dev/getting-started/readme.md).

# Introduction

**three.dev** is a comprehensive AI observability and experimentation platform. Building and iterating with Large Language Models (LLMs) is incredibly powerful, but making changes often feels like a shot in the dark. A minor tweak to a prompt or an upgrade to a new model might perfectly optimize your application, or it could completely derail your user experience.

three.dev eliminates this guesswork. The platform empowers you to test the impact of any LLM-related configuration changes quickly, reliably, and rigorously — either offline against historical traffic or live on production sessions — allowing you to move from intuition-based deployments to data-driven engineering.

## Why three.dev?

To mitigate the risks of unverified LLM configuration changes, three.dev provides an advanced experimentation framework built around *your* specific definition of success.

With three.dev, you can:

* **Observe Your Baseline**: Get deep visibility into your current LLM usage so you understand exactly how your AI configuration is performing in production today.
* **Score Production Quality Continuously**: Live scoring samples your production requests and scores each one pass/fail with your AI Judge, then groups the failures into labeled failure modes so recurring problems surface on their own.
* **Iterate Offline Before You Ship**: Replay your historical traffic through a candidate prompt, model, or parameter change. three.dev scores every output with an AI Judge calibrated by domain experts, so you can compare variants without touching production.
* **Test with Statistical Rigor**: Confidently prove when one LLM variant outperforms another. three.dev's framework uses Bayesian analysis to ensure your results are reliable.
* **Evaluate Holistic Impact**: three.dev reveals the exact impact of your changes across multiple dimensions — including accuracy, cost, and latency — so you can make the best decision for your business.
* **Experiment with Total Flexibility**: Whether you are testing a brand-new foundation model, fine-tuning parameters, or just tweaking a system prompt, you can run experiments on absolutely any LLM configuration change you want to make.
* **Protect Your Users with Dynamic Routing**: When you roll experiments out to live traffic, three.dev runs them as multi-arm bandits — evaluating multiple variants simultaneously while automatically minimizing traffic to underperforming configurations to protect your users.

Stop guessing how your AI behaves. With three.dev, you can observe, experiment, and ship with confidence.

Request access at [three.dev](https://three.dev).