Back to the changelog

v0.88.2

September 17, 2026

**A platform release, so FlowSharp only starts when it can really work:** at boot it checks that the database is up to date and says so clearly if it isn't, document reading runs in an isolated process that cannot block the service, and the PDF reading engine has been updated.

  • Security

    FlowSharp does not start on an out-of-date database

    At boot, the services check that the database schema is the one the installed version expects. If updates are missing, they do not start half-way: they log which steps are missing and the command to apply them, and start on their own as soon as the database is aligned. The readiness check used by monitoring tools now answers "not ready" in these cases, while the basic health check stays unchanged. Whoever runs the installation finds the details in the deploy kit README and in two new runbooks.

  • Improvement

    Database schema realigned once and for all

    Some indexes and internal names in the database had drifted from the ones described in the data model: a metadata-only update, with no rows rewritten and no tables locked, realigns them, and an automatic check prevents the difference from coming back in future versions.

  • Improvement

    Document reading in an isolated process

    Text extraction from PDF and DOCX files now runs in a dedicated process with its own memory limit: a malformed document may fail, but it can no longer slow down or block the service. When an extraction fails, the reason is distinct (timeout, out of memory, interruption) and shows up in the documents diagnostics.

  • Security

    PDF reading engine updated

    The component that reads PDFs uploaded into flows and the Library has been updated to a modern version: table columns no longer get glued together, but the same file, re-uploaded, may yield slightly different text. Documents already in the archive stay as they are, and a manual text correction always remains the valid version.