# three.dev

> three.dev records the LLM traffic of an AI feature, scores the requests, and groups the failures into named failure modes, ranked by severity. The team's coding agent then reads a failure mode over MCP and proposes the fix. three.dev is built by round3.

three.dev is in early access, with a Free plan and paid plans from $79 a month. Connecting an app and reading its failure modes need a three.dev account; the pages below are public. Docs pages end in `.md` for agents.

## Start here

- [Facts about three.dev](https://three.dev/facts): Name, category, what it does, who it is for, what it does not do, and who builds it, in one page.
- [Pricing](https://three.dev/pricing): Free ($0), Starter ($79 a month), and Pro ($249 a month). Every plan includes scoring and unlimited seats.
- [Docs index](https://docs.three.dev/llms.txt): Every docs page with a one-line description.
- [About](https://three.dev/about): round3, the team behind three.dev.

## Connect an app

- [Onboard with an AI agent](https://docs.three.dev/getting-started/onboard-with-an-ai-agent.md): Install the three.dev agent skills (https://github.com/round3ai/skills) in a coding agent and let it route the app's LLM calls through the proxy.
- [Sending requests](https://docs.three.dev/sending-requests/sending-requests.md): Manual setup: base URL `gate.three.dev`, a three.dev API key, and the `X-Three-Use-Case` header. The AI SDK stays the same.
- [AI providers](https://docs.three.dev/sending-requests/ai-providers-integration.md): Setup for OpenAI, Anthropic, Gemini, Azure OpenAI, OpenRouter, Concentrate, Amazon Bedrock, LiteLLM, the Vercel AI SDK, and LangChain.

## Find and fix failures

- [Live scoring](https://docs.three.dev/live-scoring/live-scoring.md): three.dev scores a sample of production requests in the background, with no added latency.
- [Failure modes](https://docs.three.dev/live-scoring/failure-modes.md): How three.dev groups failures into named failure modes.
- [For AI agents](https://docs.three.dev/for-ai-agents/for-ai-agents.md): The MCP server at `https://mcp.three.dev/mcp` gives a coding agent read access to recorded requests, failure modes, and quality metrics. The Claude Code and Codex plugins set it up.
- [API reference](https://docs.three.dev/api-reference/api-reference.md): Public API endpoints and authentication.

## Blog

- [Blog](https://three.dev/blog): Company and engineering posts from round3.
- [Hello, world. We're round3.](https://three.dev/blog/hello-world-were-round3): We started a company to answer the hardest question in production AI: is this change actually better?
- [Your use case is not in the benchmarks](https://three.dev/blog/your-use-case-is-not-in-the-benchmarks): We replayed 2,000 requests from a use case no benchmark covers and measured how seven frontier models performed across quality, latency, and cost. The winner is one no benchmark would have picked.
- [Your business can fit in a smaller model](https://three.dev/blog/your-business-can-fit-in-a-smaller-model): An introduction to the world of open-weight models for businesses wondering whether they could own their AI, where a specialized model trained on your own data can match the big ones at a fraction of the cost and latency, and where the data you already have matters more than any model you could pick.
- [Knowledge just moved out of the GPU](https://three.dev/blog/knowledge-just-moved-out-of-the-gpu): DeepSeek and Qwen just agreed on where a model's facts should live: in host RAM or on an SSD, not in GPU memory. Here's how that split changes what it costs to run a model, and what it means if you want to own a specialized one.

## Legal

- [Privacy](https://three.dev/privacy): Privacy policy.
- [Terms](https://three.dev/terms): Terms of service.
- [DPA](https://three.dev/dpa): Data processing agreement.
- [Subprocessors](https://three.dev/subprocessors): List of subprocessors.