v1.10.65 – v1.10.66 — 2026-05-20¶
- fix(training): Manual spam/ham reports now also feed the neural training corpus. Previously,
LearnSpam/LearnHamcalledrspamcwhich rspamd's neural.lua silently skips (manual scans are excluded by default unlessallow_local=trueor theANN-Trainheader is set). Switched the learn path from rspamc exec to direct HTTP POST to the rspamd controller, addingANN-Train: spam/hamheader. Combined Bayes + Neural training in a single request.