What's New in Hapi: August 2026 Update

The headline: Hapi Assistant

This update introduces Hapi Assistant — hold a hotkey, ask a question out loud, and get a hands-free answer or draft based on what's on your screen or what Hapi already remembers. No typing, no switching apps.

Read the full guide: Meet Hapi Assistant.

A cheaper AI model option

Settings → AI now offers a lower-cost cloud model alongside the existing options, for anyone using Hapi's cloud AI features. It's roughly 2.5–3x cheaper on output tokens than the cheapest option before it, and it's held to the same strict no-data-retention standard as everything else in Hapi. It's opt-in — nothing changes for you unless you pick it.

Chat is more reliable

A handful of fixes to how Hapi's assistant reads and answers:

• Reads more apps when asked about your screen. Asking "what's on my screen" used to fail on apps like VS Code, terminals, and design tools. It now falls back to reading the screen visually on any app, not just a short list.
• No more surprise email drafts. Asking a plain question no longer risks accidentally popping open a blank email draft instead of just answering you.
• Gets names right. Asking about something by its own name (a project, a product, a company) no longer answers using the name of the meeting it was mentioned in instead.
• Cleaner answers in cloud mode. Fixed a bug where a cloud-model answer could occasionally come back truncated or scrambled when it needed to look something up mid-answer.

Smarter memory and meeting prep

The biggest cluster of fixes in this update is about keeping your knowledge graph (the people, companies, and projects Hapi tracks across meetings) clean and complete:

• People and companies stop vanishing. Someone already known to Hapi but not spelled out in a meeting's summary used to get dropped from that meeting's record entirely — which also broke prep briefings and timelines for them. Fixed.
• Duplicate entities clean themselves up. The same company showing up twice — once auto-created from an email, once from being discussed by name — now merges automatically when it's an obvious match, and asks you when it isn't.
• Fewer junk entries. Generic words like "Engineering" or "brand" were sometimes saved as if they were real people or companies, cluttering your People & Entities list. That's fixed, along with a few other misclassification bugs.
• First-meeting prep cards aren't blank anymore. A briefing for someone you're meeting for the first time could come back empty even when Hapi already knew things about them. It now always shows what it knows.
• Prep cards always name their subject. No more briefings that open with a vague "that audit is still open…" — the subject is always named.
• Contradictions get flagged, not buried. If Hapi ends up holding two conflicting facts about someone, it now asks you to resolve it instead of quietly sitting on the conflict.

More reliable on lower-RAM Macs

If a meeting or voice note's processing got interrupted by memory pressure on a lower-RAM Mac, it used to stay stuck — unprocessed, unsearchable — until you relaunched the app. It now recovers automatically in the same session, for both meetings and voice notes.

Your local database also now backs itself up automatically right before any internal update, so a bad update is recoverable instead of a one-way trip.

Small fixes

• Apple Mail permission flow. Turning on Apple Mail integration now asks for permission at the right moment, and if you decline, the toggle honestly turns itself back off instead of staying "on" while doing nothing.

Questions or something not working right? Use the bug report button in the app, or reach out — we read every report.