Why Switchyard

Coding agents spend model budget fast. When the strongest model hits its limit in the middle of a task, the work stops, or you switch models and keys by hand.

Most requests in a session never needed the strongest model in the first place. Renames, summaries, lookups and commit messages run fine on a small, fast one.

Switchyard is one API key in front of every model you use. It speaks both the OpenAI and the Anthropic API formats, so your tools only change their base URL.

• Each request gets a difficulty score before it leaves.
• Hard work stays on your strongest model.
• As budgets run down, light work rides cheaper tracks.
• When every provider limit is spent, our own GPUs keep answering.
• Every reply names the model and the track that answered it.

A switchyard sorts trains onto the right track. Switchyard sorts requests onto the right model. You rank your models as tracks, and each request goes to the highest track that still has budget for it.

Track 1 · Express

Your strongest model. Architecture, hard bugs, long refactors.

Track 2 · Mainline

Everyday coding and tool calls.

Track 3 · Local

A small, fast model for light work.

Track 4 · Yard engine

Our own GPUs. No provider key needed. The last track.

There is no subscription. Routing your own provider keys is free, and your provider bills you directly with no markup from us.

The only thing you pay us for is compute on our GPUs: Yard engine usage and upgrades. You pay with credit, and credit comes from burning $YARD on Solana. $YARD is a utility token for paying for compute. It is not an investment. It has not launched yet.

The website, the docs, wallet sign-in and API keys are live. Request routing, the Yard engine and credit are being built. The changelog lists what has shipped, and nothing that hasn't.