Hello, vx

vx runs and caches a task graph, correctly, and stops there. It is a
task runner and a content-addressed cache for JavaScript monorepos, built
on Bun, with nothing hidden behind a paid cloud.

One run, then the same run again

Eight tasks across four packages, run cold, then run again with nothing
changed:

$ vx run build test --all --force result 8 tasks · 0 cached (0%) · 879ms $ vx run build test --all result 8 tasks · all cached · 2.06s saved · 26ms 

The second run is 34× faster, because every key matched and every output
was already on disk. A cold run spends its time in your tools, not in the
runner.

What is in the box, and what is not

• In core: the graph, the keys, the scheduler, the local cache and running on this machine. A workspace with no plugins runs and caches.
• Plugins: remote caches, remote execution, telemetry, CI summaries, lockfile-aware keys, AI agents over MCP. Each one sits on a documented seam, and you declare it in vx.workspace.ts.
• Never: a hosted dashboard, a daemon, or a cloud you have to buy into.

Where to go next

• Quickstart: a workspace running under vx in a few minutes, or add vx to an existing repo one package at a time.
• What vx is, and what it refuses to be: the next story.
• Benchmarks: vx, Turborepo, Nx and Vite Task on the same graph, each in its own native config.
• Features: every feature, each with a short post.

Originally published on the vx blog. vx is an MIT task runner and build cache for JS monorepos: github.com/vznjs/vx. Written with AI assistance.