VerityAi

VerityAI 4.2

Fixed

• Long-term memory could silently lose semantic search forever for any player who had facts saved before ai.embeddings was enabled — their embeddings list could never catch up in length to their facts list. Now padded/aligned correctly on load.
• Over-aggressive duplicate detection: a new, more specific fact (e.g. "likes building enormous castles") could get silently discarded as a "duplicate" of a shorter existing one ("likes building") just because one contained the other. Now only treated as a duplicate when the two are near-identical in length.
• Rare rate-limiter race condition: changing limits.max-concurrent-requests and reloading while a request was in-flight could desync the concurrency counter. The limiter now resizes in place instead of swapping in a new counter.
• The answer-arrived notification sound is now an explicit, documented setting (chat.answer-sound, default true) instead of an unconditional side effect.
• Fixed a build failure caused by Attribute.MAX_HEALTH incompatibility with older Paper 1.21.x API targets. Improved player health attribute detection for better compatibility across Paper 1.21.x versions. Fixed get_player_status function compilation and compatibility issues.

Added

• bStats integration — anonymous usage statistics (server count, versions, which AI model/features are in use). Opt out anytime via plugins/bStats/config.yml.
• Wider version compatibility — now compiled against an earlier 1.21.x Paper API target so it loads across more of the 1.21.x line, not just the exact point release it was built on.
• New default model: nvidia/nemotron-3-super-120b-a12b:free, chosen specifically for reliable tool/function-calling support, with a fallback chain (nemotron-3-ultra → ling-3.0-flash → openrouter/free) so a single model being delisted or rate-limited doesn't take Verity down.
• Refreshed README (GitHub + Modrinth) with new branding.

Notes

• No breaking config changes. New settings (chat.answer-sound) are additive with safe defaults.
• If you want the new default model on an existing server, update ai.model / ai.fallback-models in your live config.yml manually — new defaults only apply to freshly generated configs.