The news
OpenAI added an Apple Messages plugin to ChatGPT on Mac. The plugin lets the model search existing conversations, draft replies, analyze threads, and send messages after explicit user approval. It is available now through the ChatGPT Plugins interface and appears under the Public category.
Context
Until this release, ChatGPT on Mac had no native connection to the Messages app. Users who wanted the model to review conversations or suggest replies had to copy text manually. The new plugin removes that step by granting the model direct read access to the Messages database once the user installs it and grants the required permissions.
The integration arrives at a time when many Mac users already keep the ChatGPT desktop app open for other tasks. Without the plugin, any attempt to bring message history into the model required repeated copy-and-paste actions or screenshots. The new capability closes that gap by letting the model reach the same database the Messages app itself uses.
Details
The plugin requires Full Disk Access in System Settings so ChatGPT can reach the Messages store. It also requests access to contact names. OpenAI states that sending any message still needs a one-time confirmation by default and warns users against turning on persistent approval.
OpenAI listed four example prompts to demonstrate the integration. One prompt asks the model to check the calendar and reply to a named contact with available dinner times. Another asks it to suggest follow-ups based on messages from the previous day. A third directs the model to locate birthdays mentioned in messages and add them to the calendar. The fourth asks it to identify potential spam for deletion. The summary also notes support for handling RCS messages.
Installation happens inside the ChatGPT app under the Plugins section. No additional software download is required beyond enabling the plugin and approving the system-level permissions.
The Full Disk Access requirement stems from the location of the Messages database on macOS. That database sits inside a protected folder that standard app sandboxes cannot reach. Granting the permission lets ChatGPT read message contents, attachments metadata, and contact information stored alongside the conversations. OpenAI has not published the exact database schema it accesses, but the listed prompts imply the model can parse timestamps, sender names, and message text.
Each prompt example shows a distinct workflow. The calendar-and-dinner prompt combines message history with calendar data the user has already shared elsewhere. The follow-up prompt operates only on recent threads. The birthday prompt requires the model to scan older messages for date references and then trigger a calendar entry. The spam prompt focuses on pattern detection across multiple conversations. Support for RCS messages extends the same capabilities to conversations that use the newer protocol rather than classic SMS.
Why it matters
The change gives ChatGPT a direct line into a user’s primary messaging channel on the Mac. For people who already run ChatGPT on their desktop, the plugin removes friction around message review and reply drafting. At the same time it introduces a new permission surface that touches the full message history and contact list, which increases the value of the explicit approval step OpenAI has kept in place. Users who value quick assistance with scheduling or spam cleanup now have a concrete tool, while those concerned about data access can simply leave the plugin disabled.
The permission model itself reveals a tension that will recur with other desktop AI features. Full Disk Access is an all-or-nothing gate on macOS; once granted, the app can read any file the user can reach. OpenAI’s decision to keep message sending behind a per-message confirmation reduces one class of risk, yet the read path still exposes years of personal correspondence. People who treat their Messages database as a lightweight archive of both business and personal exchanges now face a clearer choice: accept the convenience or keep the model at arm’s length.
Over time the plugin may shift expectations about what desktop AI assistants can do with local data. If the pattern holds, future plugins will request similar broad access to other apps. Users who adopt early will test whether the productivity gains outweigh the expanded attack surface and the possibility of future policy changes around stored message retention.
---
Sources:
{"word_count": 682, "sources_used": 1, "expanded_sections": ["context", "details", "why_it_matters"]}
No comments yet