Fixed
- Re-embedding to a custom provider no longer breaks the store.
iai-mcp migrate --reembed-to-configured-providerrebuilt the records table
from a schema that dropped the primary-key constraint, so records captured
afterward stored a null internal label and the daemon crash-looped on the next
restart with memory reported unreachable. The migration now preserves the
table's canonical constraints; a null label left by an affected earlier run is
repaired automatically on startup; and a genuinely null label now fails with a
clear, actionable error instead of an opaque one. If you hit this on 2.4.0 or
2.3.1, upgrading and restarting is enough — no manual repair needed. doctorno longer reports a false failure while the daemon is winding down.
Check (e) rejected theTRANSITIONINGstate the daemon legitimately writes on
the way to drowsy, showing a spurious failed check on a healthy install. It now
accepts that state.