I’ve contemplated building my own CI tool (with a local runner) and the thing is if you assume “write a pipeline that runs locally but also on push”, then the feature depth is mostly about queuing, analyzing output, and often left off, but IMO important, charting telemetry about the build history.

Most of these are off the shelf, at least in some programming languages. It’s the integrations and the overmanagement where a lot of the weight is.