Yeah it still uses context but way more efficiently, instead of injecting a 50KB context.md every time, Recall searches 10k memories and only injects the top 5 relevant ones (maybe 2KB), so you can store way more total knowledge.