AsgineAI changelog

• Increased the completion budget so ordinary conversations no longer stop after only a few paragraphs.
• Character replies now appear as soon as writing finishes while memory, relationship and scene analysis continues safely in the background.
• Stalled or unreachable generation requests now switch promptly to the backup provider.
• Short retry timing replaces long retry chains during temporary provider outages.
• Story responses receive a larger completion budget to reduce cut-off scenes.

• Important long-term preferences, boundaries and ongoing duties now remain available in long conversations.
• Story memory is selected using the current scene instead of unrelated recency alone.
• Continue now advances from the previous moment and automatically rejects accidental repetitions.
• Empty or malformed provider responses retry automatically before switching to the backup provider.
• Failed drafts are not added to the story.

• Added five spaced retries for temporary text-generation capacity errors.
• If the primary provider remains unavailable, generation can continue through DeepSeek directly, preserving the selected Fast or Pro writing mode.
• Existing streams are never restarted after an answer begins.
• Internal cost estimates now account for the backup provider's pricing.

Branch continuity

• Rewinding or restoring a branch now brings back its story board, stats, scene context, appearance, relationships, meters, cast and character evolution.
• NPC and cast state stays consistent when switching timelines.

Continue & Regenerate

• Continue and Regenerate now save the correct state snapshot for both Classic and Story chats.
• Regenerated messages can no longer leave behind active memories from discarded replies.

Memory reliability

• Memory updates now use a persistent retry queue with crash recovery, protected worker locks and automatic backoff.