CharMemory

CharMemory (bal-spec) automatically extracts structured character memory from a conversation into the Data Bank and recalls it via vector search at generation time; it coexists with lorebook-based memory.

How it works: as you chat, CharMemory sends conversation snippets to an LLM (your choice of provider — including free options like Pollinations) and asks it to extract key facts, relationship changes, and events. These are saved as structured markdown files in SillyTavern’s Data Bank. When the AI generates a new reply, Vector Storage finds the most relevant past memories and injects them into the prompt.

Injection Viewer lets you see exactly which memories are being recalled for each message, making it easy to debug when the AI “forgets” something. If a memory isn’t being retrieved, you can tune the embedding provider, relevance threshold, or query settings.

Group chat support means each character in a group gets their own memory file, so the system tracks what each individual character knows and experiences.

Important: back up your Data Bank before bulk operations like “Consolidate” or “Clear All” — these modify or delete files and can’t always be undone.