Amazon Bee Frontier and Eclipse: Two AI Recorders at $75 and $100, Plus a $12 Monthly Plan

Amazon has launched the first new Bee hardware since it bought the startup in 2025: the Bee Frontier, a $74.99 wearable you can wear on the wrist or pin to clothing, and the Bee Eclipse, a $99.99 credit-card-sized recorder that sticks magnetically to the back of a phone. Both listen to your day, turn it into summaries and to-dos, and can plug into Gmail, Google Calendar, Outlook, Amazon and Alexa+. A free tier keeps 14 days of memory; Bee Pro costs $11.99 a month, per TechCrunch and Bloomberg's report summarized on Techmeme.

That is the news. The more useful question for a reader is whether an always-listening recorder from a company that also sells Echo and Ring belongs in your life, and what you are really paying for. This post lays out the confirmed specs, the pricing math, the privacy claims and where they stop, and how Bee compares with the other AI recorders on the market.

TL;DR: Bee Frontier vs Bee Eclipse

What exactly did Amazon announce?

According to TechCrunch's launch report, both devices record everyday conversations and convert them into summaries, to-do lists, calendar invites and insights. They connect to Gmail, Google Calendar and Microsoft Outlook, and can draft emails and set up meetings. They also link to Amazon.com, Alexa+ and Apple HealthKit, so you can place an Amazon order or ask Alexa a question hands-free.

Hardware is deliberately minimal. Each device has a single power button and an indicator light that shows when a conversation is being summarized. The Frontier is the more rugged piece: IP67 water resistance, USB-C charging, dual microphones with noise filtering, and a claimed battery life of up to seven days. The Eclipse trades that wearability for convenience, riding on the back of your phone like a MagSafe card holder.

Amazon's earlier newsroom post from Bee's co-founder, Building Bee at Amazon, describes the same privacy approach (real-time processing, no stored audio, transcripts only accessible to the customer) but predates these devices; the launch specs here come from TechCrunch. We could not find an independent teardown or battery test yet, so the seven-day figure and the microphone claims are Amazon's.

How much does Bee really cost?

The sticker prices are low. At $74.99 the Frontier undercuts the competitors TechCrunch lists: the Plaud Note at $159, the Plaud Note Pro at $189, and Pocket AI at $129. The Eclipse at $99.99 still sits below all three.

The subscription is where the economics change. The free tier is genuinely usable for light users: summaries and recaps with a two-week memory. But the features that make an assistant feel like a memory, such as unlimited summaries, suggested to-dos, daily memories and a chat assistant that can answer questions about your past conversations, sit behind Bee Pro.

Those totals are our arithmetic from the published prices and exclude tax. They matter because the cheapest hardware in the category is not the cheapest to own if you want the paid features. If you only want a searchable record of meetings, one of the pricier one-time-purchase recorders can come out ahead over two years. If you want Alexa and shopping hooks, only Bee has them.

What does Amazon say about privacy?

Privacy is the first objection anyone raises to a wearable that listens all day, and Amazon addresses it directly. Per TechCrunch's summary of the launch:

• Raw audio is not stored. It is processed in real time and then deleted.
• Transcripts are encrypted with keys the device creates and controls, so Amazon employees cannot access them unless the user intentionally shares them.
• Connecting to Alexa is optional. It enables Voice ID so an Echo can recognize when the user is speaking.
• Users can delete transcripts and summaries at any time.

These are meaningful design choices, if they hold. Deleting audio after transcription limits what can leak or be subpoenaed. Device-held keys would limit internal access. But all of this is Amazon's description of its own system. Independent reviewers have pushed back since Amazon first bought Bee: BGR argued the privacy protections look real on paper but incomplete given Amazon's history with Echo and Ring, and flagged bystander recording and a hard-to-see LED when worn on the wrist. Those points do not prove a problem, but they are the right questions to ask.

Two gaps worth knowing about:

1. Bystanders. The person wearing a recorder consents; the people talking to them may not. An indicator light signals that summarization is happening, but it is not consent, and recording laws vary by state and country.
2. Cloud processing. Real-time processing means your conversation leaves the device in some form, even if the audio is deleted afterwards. Transcripts and summaries still live somewhere, and a breach of that store is a different risk from a breach of raw audio.

For a broader look at how agent and assistant makers word their privacy promises, see our analysis of AI agent privacy promises from Meta and OpenAI.

Why does Amazon want this product?

Bee was a small startup making a wrist-worn AI recorder before Amazon acquired it in 2025. These are the first new products since the deal, and the integration list tells you the strategy: Alexa+, Amazon.com purchasing and Apple HealthKit. A device that hears what you agree to buy, plan or remember is a natural input for an assistant that can then act on it.

That puts Amazon in a crowded race for the personal-AI device. Meta has been pushing its own assistant and accessories; our Meta Muse gadgets and SDK post covers that side, and our Meta Muse dossier investigation shows how quickly privacy questions follow assistants that remember. On the other end of the spectrum, local-first hardware is growing: the Oki Home local AI computer and the Verge local AI diary are examples of people choosing to keep models and memory at home.

Who should buy it, and who should wait?

Reasonable buyers: people who already live in Alexa, Gmail and Outlook and lose action items after meetings; field workers who cannot take notes; anyone who wants a cheap trial of an AI recorder before paying for a premium one. The free tier lets you test for a two-week memory window at the cost of the hardware.

Wait or pass if: you work with confidential material, regulated data or clients who have not agreed to be recorded; you live where all-party consent is required and do not want to manage it; or you object to a retailer holding a running record of your conversations, even in summary form.

A practical checklist before turning one on:

1. Decide where it stays off: therapy, legal calls, HR meetings, anything under NDA.
2. Tell people. A sentence at the start of a conversation solves most consent problems.
3. Skip optional integrations at first. Connect email and calendar only after you have seen what the summaries contain.
4. Delete test transcripts and check that deletion works in both the mobile and desktop apps.
5. Keep Alexa linkage off unless you need Voice ID on an Echo.

What is not confirmed yet

• Eclipse release date. Reports say "soon after" the Frontier but give no date.
• Battery and accuracy. The seven-day battery and microphone quality are Amazon figures. Wait for independent reviews.
• Regional availability. The coverage we read is US-focused; we found no international rollout details.
• Data retention beyond 14 days on the free tier. Amazon says memory is 14 days; how long underlying summaries persist after that is something to check in the privacy settings.
• Audits. We found no third-party audit of the encryption and deletion claims.

What this means for builders

AI wearables are becoming a distribution channel for assistants. If you build agents, a steady stream of recorded context, such as meeting notes, to-dos and calendar events, is exactly the data layer assistants need. It also sets expectations for users: they will soon expect any assistant to remember what was said yesterday. For the infrastructure angle on running that memory locally rather than in a vendor cloud, see our piece on Windows hybrid local-and-cloud agents. And if you are deploying agents that act on recorded data, AgentBeam is the agent security platform from the explainx.ai team that stops AI agents before they take dangerous actions.

Related reading

• Meta Muse gadgets and SDK
• Meta Muse dossier investigation and privacy
• AI agent privacy promises compared
• Oki Home local AI computer
• Verge local AI diary
• Windows hybrid intelligence

Specs, prices and availability are accurate as of October 15, 2026 and come from Amazon's announcement and press coverage. Confirm on Amazon before buying.