Lagoon readies local-first AI chat workspace

Lagoon’s real pitch isn’t “AI chat” so much as a transparent memory stack for long-form fiction, which is where most chat apps fall apart.

• –On-device `all-MiniLM-L6-v2` embeddings keep the memory loop private and free of embedding API costs.
• –Runtime controls for top-k, similarity threshold, chunk size, and token budget are the right knobs for debugging retrieval quality.
• –The style overseer is a strong differentiator because it fixes prose after generation instead of relying on a perfect prompt.
• –BYOK support across hosted and local providers makes the stack flexible for both tinkering and heavier usage.
• –The fiction-first workflow should still translate well to RPGs and other long-context collaborative chat use cases.