How a stack is built to keep working when a vendor goes dark: the cutoff test, an EU-first model, local runtime tiers, and self-hosted messaging.
The bot my agents reported through worked, which is why the dependency was worth breaking: the channel was mine to use and never mine to keep. This piece follows the move to a self-hosted, non-federated, registration-closed Matrix homeserver, chosen for sovereignty rather than to fix a failure.
Your dictation audio either stays on the device or it does not, and that line picked the stack. This piece follows the on-device choice once audio egress became the one disqualifying criterion, and what ruling out the cloud apps gave up without pricing.
Moving a household off the US cloud is not one migration but four decisions, and the pieces that do not move cleanly are the risk, not data loss. This piece walks the mail, docs, and calendar move that ran, the accounts left open on purpose included.
Where a task runs decides who can reach its data. This piece lays out the local-first default across every profile, the scoped profiles that keep a cloud path, the guardrail that enforces the local tier on sensitive tasks, and what a 24 GB machine will actually hold.
A model per task tier is costly to operate and no safer against a discontinuation. This piece shows how one EU default primary, tuned per task by a binary reasoning toggle and backed by a two-provider fallback family, replaced the whole tier instead.
Show operators and site owners how a four-site WordPress move off managed hosting was run as a blue-green DNS flip, so the move carried a reversible, zero-data-loss rollback instead of a one-shot cutover.
Give engineers and technical founders choosing AI infrastructure one ranking test, whether each dependency keeps working if its vendor cuts service to them, so migration priority is set by cutoff survival rather than by the jurisdiction on a map.
