Ill-advised, and running in production.
yah is a software development stack for SaaS free cloud apps — agent-chat, productivity board/docs, verified loops, and uh "the cloud". Every part of it is something you should have outsourced.
confer · codify · ship
Five or more disjointed services, each with its own flaky or undiscoverable MCP, collapsed into one process. The last twenty-five years say don't do this. Two well-worn loops carry most of the traffic; everything else is a back-edge — sometimes friction, sometimes the most valuable signal in the system.
ChatTicketsDocsGatesReleasesNone of these concepts are new. They're usually five apps that, at best, each grew an MCP. Here it's one app, and the agents hold the keys to every step — board, docs, files, test runs, queries, machines, services, releases.
You could have rented the last mile. We wrote it instead.
That last mile turned out to be three programs. One provisions and reconciles machines you already own; one renders and publishes the sites that run on them; one decides whether any of it was allowed to ship. They live in their own repos, under their own names, and the desktop app drives all three.
You write TOML describing machines you already rent or own; yah cloud makes the live fleet match it — provision, deploy, ingress, rollout, teardown.
One web project, three modes — static, server-rendered, or SPA shell — chosen per route. It built the page you're reading.
Pipelines and step DAGs over your task runner. It decides what passed, and hands the artifact to the deploy.
This page is one of those workloads: kind = "mesofact-static", built by bun run build, published by the cloud reconciler. The bytes still come off a CDN rather than one of our own machines — that migration is half finished, and saying so is cheaper than pretending otherwise.
One line, then the cursor blinks.
The macOS build is signed and notarized; Linux and Windows ship as verified tarballs, sha256-checked by the installer on every platform. The CLI is the same binary the desktop app drives — no two implementations to keep in sync. It's built by a small team rather than a trillion-dollar lab, which is either the appeal or the objection, depending on you.