Solid comparison. One nuance I would push back on: "Typesense keeps the entire index in RAM" can be misleading for self-hosters planning capacity. If your dataset grows beyond available RAM, Typesense degrades hard — there is no graceful spill-to-disk like Elasticsearch. For homelab setups with 8-16GB RAM, that ceiling matters. Have you tested what happens when the index approaches 80% of available memory?