A new application called Hemory allows AI assistants to access real-time conversations by listening to and recording interactions, then providing this information to AI models such as Claude, Codex, or Gemini. The app, developed by the Californian company Beautiful ONES, was first launched on iOS and Apple Watch on August 1st, with several updates following in the weeks after. It operates by continuously listening to the user's environment, either manually or on a schedule, and then transcribing and summarizing these interactions into a structured memory database. This database can be accessed in read-only mode by compatible AI agents through a feature called MCP. Hemory offers a free trial with 10 hours of listening time, and subscription options include $10 per month for 20 hours, $20 for 60 hours, or $50 for unlimited listening time. The app uses voice activity detection, which means silences, pauses, and background noises are not counted toward the listening time. The service claims that raw audio is not stored on AWS servers located in the United States, but rather processed and then destroyed, with only transcriptions and voice prints being stored until they are deleted. The company also states that these data are not used to train models, although some functions may transmit them to Anthropic, OpenAI, or Microsoft Azure. Hemory also promises a self-hostable version, which would allow users to maintain more control over their digital memory, although this feature is not yet available and is described as "coming soon." The app includes features such as transcription, identification of interlocutors, and access through MCP. However, the service reminds users that they must obtain necessary consents and respect legislation when recording conversations, as the recordings may include colleagues, loved ones, or other interlocutors. This raises privacy concerns, as the app listens to almost everything, potentially recording and preserving traces of others' statements.