> 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** finds out how your AI agent fails, so you can fix it. It reads your production LLM traffic, scores a sample of it, and names the specific ways your agent fails. You learn what to fix from your own traffic, not from user complaints, and you never write or maintain an eval suite.

You route your LLM calls through the three.dev proxy with a few headers. The proxy is built to be fast and highly available. It forwards each request to your AI provider, and records it only after the response is on its way back to your app, so recording adds almost no latency. Requests are grouped by use case and session. A coding agent can do the setup for you with one prompt.

## Why three.dev?

With three.dev, you can:

* **See your real traffic**: Read the LLM calls and conversations your application makes in production, with their model, latency, cost, and tags.
* **Find the issues in production**: Live scoring samples your production requests and finds the issues in them, so quality on live traffic is a number you can track.
* **Know what keeps going wrong**: three.dev names the recurring problems as failure modes, rated by severity, so you know what to fix first.
* **Fix it with your coding agent**: Your coding agent reads the failure modes and the requests behind them through the three.dev MCP server, and proposes a change to your code.
* **Slice by what matters to you**: Tag requests and sessions with your own attributes, such as environment, plan, or product line, then filter the Overview, Requests, and Sessions by them.

Sign up free at [three.dev](https://three.dev?ref=docs-introduction), then [onboard with an AI agent](https://docs.three.dev/getting-started/onboard-with-an-ai-agent.md).
