Ship one app, not the monorepo
vx prune copies a project and its dependencies with a pruned lockfile, for a Docker build.
A Docker image for one app should not install the whole monorepo’s dependencies. vx prune @demo/web --docker writes just the projects it needs and a lockfile cut down to them.
Copy json/ first and install, then full/, and the install layer caches until a dependency changes.
For the image that only runs the app, --production also leaves out workspace packages that only dev dependencies pull in.
// what it prints
$ vx prune @demo/web --docker
vx prune: 2 projects → out (json/ + full/), bun.lock prunedMore in Plugins
A pipeline with seams
Fourteen seams, and core applies no plugin by default.
Remote cache and execution
Distributed builds over Bazel’s open Remote Execution API.
OpenTelemetry
Every run as OTLP traces, metrics and logs, with no SDK.
vx mcp for coding agents
Give your agent the build’s memory over MCP.
Longest chain first
The slow task stops starting last.
This machine is always the floor
Running and caching locally are built in, so a declined plugin hands work back, never drops it.
What the scheduler learned
vx history shows each task’s typical time, peak memory and CPU from past runs.
Code around the run
A plugin can start something before the run and clean up after, within a deadline.
Hosted remote servers, Bazel-style
TLS, mutual TLS and auth headers connect vx to servers such as BuildBuddy.
Remote runs that never repeat
A remote action that already ran replays its outputs and log, without a worker.
A wedged server is just a miss
Downloads are verified and calls have deadlines, so a bad server never hangs a run.
node_modules for stateless workers
Run the install as a remote action, so every worker has node_modules without a shared disk.
Watch CI while it runs
Spans and metrics stream as tasks end, so a dashboard follows the run live.
Tasks packed by memory
The history plugin admits tasks by the peak memory they used before, so a run never runs out.