Schema Drift Watch
Compares declared schema against live migrations. Detects narrowed columns, type renames, and orphan migrations that AI builders silently introduce on feature branches.
Every SwarmWarn detector reads a single surface deterministically, then explains itself in plain language. AI is allowed to describe a finding — never to invent one.
Compares declared schema against live migrations. Detects narrowed columns, type renames, and orphan migrations that AI builders silently introduce on feature branches.
Measures bundle + dependency weight against your target host ceiling on every push. Knows the ceiling before the platform does.
When Watch names a wall, Recovery returns the minimum-change recovery path: which dependencies to drop, which routes to defer, and the rollback build that's already clean.
Reads the entire deployment surface — environment variables, edge runtime constraints, route segment compatibility — and names exactly which limit will block the candidate build.
Protects high-value surfaces — auth flows, billing, schema files — from quiet AI-builder deletion. Treats unannounced erasure as a named wall, not a clean diff.
Every metric is timestamped, sourced, and chained to its build. Your board sees attestations, not vibes — and the chain cannot be retroactively edited.
Structured fix plans: the named wall, the minimum recovery diff, the rollback identifier, and the exact rule id behind each finding — every report PR-ready.
SwarmWarn never depends on a single model. Deterministic rules execute first; OpenRouter and MiMo are wired for failover so verdicts return even when a provider is down.
Detections must arrive before you promote. A pager at 3am is not a verdict; a named wall before promotion is. We analyze static builds before the container is locked, protecting you from breaking production pipelines.
Container Sizing is where most teams find their first wall. Start there.