Howzer v2.5.1-DE: in-house sentiment model, report delivery, more hardening
A maintenance release that strengthens three areas: an in-house German sentiment ML model, automated report delivery by email, and further robustness hardening for self-hosted operation.
By Howzer Team, Product
What's in v2.5.1-DE
v2.5.1-DE builds directly on the v2.5.0 hardening work. It is a maintenance release — no new analysis dimensions — that closes three gaps: sentiment now runs on an in-house German model instead of an upstream one, scheduled reports are delivered by email out of the box, and the persistence and engine layers were hardened further for high-volume, self-hosted operation.
In-house sentiment model
Sentiment analysis moves to a dedicated German ML model, trained and calibrated on German feedback like the other custom Howzer models. It is enabled by default and can be toggled with the SENTIMENT_USE_ML flag, which falls back to the previous behaviour when disabled.
Report delivery by email
- The consolidated report engine now sends scheduled reports (daily and weekly) as email with the report attached, driven by the single delivery workflow.
- Recipients and sender are configured with REPORT_RECIPIENTS and REPORT_SENDER; without them the delivery step is deliberately skipped rather than failing.
Robustness hardening
- Persistence and engine paths were hardened for large batches (10k+ findings) so runs stay correct and honest under load.
- Database migrations are at head 016 — the CLI installer applies them automatically on install and upgrade.
- n8n runs with N8N_BLOCK_ENV_ACCESS_IN_NODE enabled, so workflow code cannot read the container's secrets from the environment.