Switchyard

Software / Engineering Teams · Developer tooling

One gateway every tool points at. It meters every call's tokens and cost by tool, engineer and team, routes each call to the cheapest model that can do it, and holds teams to their budgets.

[ fig. 1 ]The Switchyard launch film: five coding tools send every call to the biggest model, Switchyard routes each one onto a frontier, standard or small track, and its dashboard shows a month of spend, $6,005 saved, and team budgets live. Recording, 0:11–0:14.

The problem

Engineering teams run five AI coding tools, and every one sends every call to the biggest model, autocomplete included. Nobody could say which tool, engineer or team the bill came from.

Before: Every call on the biggest model
After: Each call on the right one

How it’s built

Every coding tool

Claude Code

Cursor

Codex

Devin

Antigravity

SwitchyardExample callCursor · autocomplete · 800 tok

IdentifyMeterRouteBudget

Rule: routine task → small

Metered by tool, engineer and team

Model tiers

Frontierclaude-opus-5-5 · gpt-5gpt-5 → —

Standardclaude-sonnet-5-5 · gpt-5-minigpt-5 → —

Smallclaude-haiku-4-5 · gpt-5-nanogpt-5 → —

[ fig. 2 ]A drawing of the system, not a screen. Swipe sideways to see all of it.

In numbers

$72k: yearly saving
39%: cheaper than unrouted
158k: calls routed monthly

For a team of 24 engineers making about 158,000 calls a month, at standard model prices. Routing small edits to small models too saves $6,005 (38.8%) a month; the default rules save $5,051 (32.6%).