Fine-tuning bakes yesterday's data into frozen weights: expensive to update, impossible to audit, unable to forget, and blind to permissions. Heidi keeps knowledge in a live memory the model reads at answer time, current, sourced, and deletable.
The operational problems are decisive. A fine-tuned model cannot tell you where an answer came from, cannot honor one person's access being different from another's, and cannot delete what a customer asks you to delete.
Context beats weights for company knowledge: it updates the moment something changes and improves automatically as frontier models improve, because Heidi rides the best available model instead of freezing one.