The context tax is what you pay every time an AI has to rediscover what your company already knows: tokens burned re-reading repos and documents, and answers that are still guesses. Heidi ends it by making context a one-call lookup.
The clearest case is coding agents: brilliant, and new to your company every morning. Without memory they crawl the codebase to guess at 'why' questions; with Heidi they ask once and get the institutional answer with receipts.
The math is dramatic, hundreds of tokens instead of hundreds of thousands, but the quality shift matters more: an answer with sources beats a well-informed guess.