Voxta Microsoft Semantic Kernel
Vector-based local memory using sentence-transformer embeddings.
Microsoft Semantic Kernel provides a vector-based memory backend in Voxta. It uses local sentence-transformer embeddings (no API key, no cloud) to do semantic lookups — "the cat" can match an entry about "felines" because the embeddings are similar, not because the words overlap.
Compared to Simple Memory it's slower (vector lookup costs more than keyword match) and uses more disk/memory (stores embeddings), but the retrieval quality is meaningfully better for scenarios with lots of stored facts.
Setup
Add the service
Manage Services → + Add Services → Microsoft Semantic Kernel → Add. Voxta downloads the sentence-transformer model on first use.
Configure (optional)
Default settings work for most chats. Advanced options are exposed for embedding model selection and similarity thresholds.
Note
The Semantic Kernel framework also has broader uses (tool orchestration, agentic workflows) — this Voxta module specifically uses its memory subsystem. For general SK integration see Microsoft's Semantic Kernel docs.