Hemory is a tool that connects "real memories" to AI agents, its name combining Hear and Memory. Its core idea is: through always-on listening, it captures conversations and events you experience firsthand, precipitates this content into private, searchable memories, and then connects to your agent via MCP (Model Context Protocol), so that AI no longer relies only on temporary input, but answers questions and generates content based on "things that actually happened."
In terms of main capabilities, Hemory provides continuous listening and memory precipitation; all content you have experienced enters a private, searchable memory library. It connects to mainstream agents via MCP; for example, in the command line you can use codex mcp add hemory or claude mcp add hemory to complete the integration, and it supports more standard MCP clients. After connecting, the agent can call the search_memory tool to retrieve memories. The official examples show multiple uses: for example, when following up on "What did Daniel promise me at last Tuesday's sync meeting," the agent can retrieve the relevant conversation record via hemory search_memory (MCP) and give a conclusion; it can then also generate a follow-up email draft, such as Write(follow-up.md), totaling 12 lines and ready to send.
Usage is divided into two steps: first enable listening in Hemory, then connect Hemory to your agent via MCP. Listening modes provide manual mode (enabled only when you need it, with privacy boundaries under your control) and scheduled mode (for example, weekdays 9:00–18:00, with a reminder to enable at the set time). After connecting, various generation tasks can be performed based on memories. The official examples list a monthly work report demo page, a diary at 22:00 every night, and a smart ring PRD draft, showing that the same memory source material can support multiple documents or question-answering tasks.
Applicable scenarios include: personal daily memory and review, work sync records and follow-ups, precipitation of customer interviews and brainstorming, and quickly turning scattered conversations into documents such as reports and PRDs.
Regarding privacy and limitations, Hemory claims zero cloud retention of audio: audio is processed as a stream and destroyed immediately after processing is complete, and is not stored in the cloud; raw audio is stored only on the listening device and is not synced across devices. The official information also mentions that a self-host option is coming soon, allowing it to run entirely on your own infrastructure. On billing, it mentions a one-time, non-renewing, VAD (voice activity detection)-based quota mechanism; VAD counts only the time when someone is actually speaking, and silence, gaps, and background noise are not counted toward the quota, with a 24-hour/day cap and fair use limits. Regarding the specific price amount, the materials do not currently provide it. Supported platforms include macOS, Windows, Linux, and Web, with self-hosting coming soon.
Please note that the above features, privacy commitments, platform, and billing descriptions all come from the official introduction text, and actual performance may change with versions; the specific price and complete limitation terms are not currently provided in the materials.
