Halo (Mac)

Most AI tools that watch your screen, transcribe your calls, and autocomplete your writing do it by shipping everything to a cloud model you have no visibility into. Halo takes the opposite position: it runs on your Mac, learns from what crosses your screen and your calls, and surfaces that context — drafts, summaries, recalled details — through a persistent presence in the notch. The core loop is voice activation via ⌘⇧M or ‘Hey Halo’, contextual ghost-text autocomplete in any app, one-shortcut call recording with speaker-labeled transcripts, and plain-language recall of past screen content and conversations.

The differentiating feature is the combination of persistent local memory with autonomous monitoring. You describe a standing job — watch this inbox, track this repo, alert me when this price moves — and Halo runs it on your machine while you work. Critically, the vendor states it notifies but never sends, buys, or deletes without your approval, so you stay in the loop on every consequential action. For teams where compliance or confidentiality rules out cloud agents, that architecture matters.

Halo fits a specific profile: a Mac user who wants unified voice control, meeting notes, contextual writing assist, and lightweight agent automation without surrendering data to a third-party server. It breaks, or simply doesn’t exist, the moment you need cross-platform coverage, an API to connect Halo’s memory to other systems, or agents that execute autonomously without a confirmation step. Teams building shared automation pipelines will hit that wall fast.

On the integration side, the vendor describes built-in connections to Calendar, Mail, and Contacts, plus support for any MCP server — so you can point Halo at additional tools via the Model Context Protocol and it will use them within the same on-device loop. There is no public API, which means Halo’s context and actions stay local to the machine and are not programmable from external systems.