Answers from the documentation; checking which model is on duty
Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
New AI wearables add voice, memory, and companion features - OpenSmartRoute
New AI wearables add voice, memory, and companion features
TechCrunch reports on new AI wearables from Amazon, Friend, Meta, and others. These devices now record audio, summarize conversations, and act as personal companions.
Key points
Amazon launched the Bee Frontier for $74.99 and Bee Eclipse for $99.99.
Friend 2.0 includes a built-in speaker for two-way communication.
Limitless offers 10 hours of AI features for $99 with unlimited options for $29.
The Stream ring is expected to launch in summer at $299.
Why it matters: Teams can evaluate how these devices handle audio transcription, context retention, and privacy controls for their agents.
By OpenSmartRoute editorial · written through the router by writer-small
From TechCrunch AI - “The hottest AI wearables and gadgets you should know about”
Introduction - AI wearables are entering the market as productivity and companion tools.
A new wave of artificial intelligence devices is arriving on the market. These gadgets aim to integrate intelligence into daily life like never before. Some devices act as productivity tools for work and study. Others claim to serve as friendly companions for personal thoughts. Even major companies like OpenAI are developing compact AI companion hardware.
These products range from necklaces and rings to wristbands and portable units. They record audio, summarize conversations, and offer personalized advice. Users can interact with these devices directly through voice commands. The technology replaces traditional note-taking apps and voice assistants.
Many of these devices listen constantly to capture context. They learn user routines to create tailored reminders and notes. Privacy features like mute buttons allow users to control data collection. The companion apps often require a monthly subscription for full access.
This shift marks a move from cloud-based AI to edge computing. Devices now process data locally before sending summaries to the cloud. This change affects how engineers design models and how managers evaluate costs.
Amazon's Bee devices - New models and pricing details for the wearable line.
Amazon acquired the Bee startup in July 2025. The company launched new products earlier this month to refresh its lineup. The original Bee pendant costs $49.99 and clips to clothing or acts like a fitness band. It records audio and learns user preferences to generate reminders.
Amazon introduced two new models to compete with rivals. The Bee Frontier costs $74.99 and wears on the wrist or clips to clothes. The Bee Eclipse costs $99.99 and attaches magnetically to a smartphone back. Both devices offer the same core functionality as the previous generation.
Users can connect Bee to Alexa for voice commands. This integration allows Bee to answer questions using Amazon's ecosystem. The companion app is currently available only on iOS devices. It requires a $19 monthly subscription for direct interaction.
The app provides key takeaways from the user's day. It also offers chronological transcripts of all conversations. This feature helps users review their interactions later. Bee positions itself as an affordable alternative to competitors like Plaud.
Block launched Dessa, a voice model that runs on consumer GPUs and sounds human. The company invited safety researchers to test it before public release.
Native audio models process sound directly without converting it to text first. This method preserves timing, tone, and speaker cues that text transcripts lose.
Amazon's acquisition signals a push into the personal AI hardware market. The company leverages its existing voice assistant infrastructure. This strategy lowers the barrier for consumers compared to standalone devices.
Friend and Limitless - How these pendants function as emotional support and meeting notes.
Friend is a white pendant priced at $129 that hangs around the neck. It functions as an emotional support companion by recognizing tone and mood. Users can chat with it as if it were a real friend. The device connects to a phone via Bluetooth and listens constantly.
Friend sends proactive messages to users based on detected emotions. For example, it might wish a user good luck before an interview. The company faced criticism over a subway ad campaign in New York City. People vandalized the ads, writing messages like "surveillance capitalism."
To stay relevant, Friend unveiled a major product update recently. The new version, called Friend 2.0, includes a built-in speaker. This allows the device to communicate directly and express a distinctive personality.
Limitless, formerly known as Rewind, is another conversation-recording pendant. It costs $99 and continuously listens to meetings and calls. With consent, it transcribes conversations into searchable and summarized knowledge. This feature is ideal for professionals like journalists who need to recall details.
The Limitless companion app includes 10 hours of AI features per month. Users can unlock unlimited features for $29 per month. This subscription model covers transcription and summarization tasks.
Both devices face privacy concerns regarding constant listening. Engineers must evaluate how these devices handle sensitive data. Managers need to consider the cost of subscriptions versus the value of insights.
Meta and Nirva - Upcoming hardware and the focus on personal growth.
Meta introduced its first AI wearable in September. The device offers a new way for users to access its personal AI assistant, Muse. It has not yet shipped to consumers, but details are available. The hardware is expected to feature Jolly, a playful digital avatar.
The wearable is pocket-sized and can be carried or attached to a keychain. It includes a miniature display showing the avatar's face. Users interact with Muse through voice commands and task completion. This setup replaces the need for a separate phone for simple tasks.
Nirva is a startup founded by former Meta Reality Labs employees. They are building an AI wearable focused on personal growth. The device listens to speech and uses AI to turn experiences into a personal journal. This helps users understand their emotions, communication style, and habits.
Through its companion app, users explore patterns in their behavior. They gain insights into how they communicate and find ways to improve. Nirva plans to open orders in November. Shipments to early users are targeted for the end of the year.
The Nirva device will cost $249 when it launches. This price point positions it as a premium personal growth tool. It differs from productivity-focused wearables by emphasizing self-reflection.
Omi and Plaud - Specific capabilities for conversation summarization and note-taking.
Omi is a device priced at $89 that answers questions and summarizes conversations. It creates to-do lists and helps schedule meetings automatically. Omi is constantly listening and runs conversations through ChatGPT. This allows it to remember context and offer personalized advice.
Users can wear Omi as a necklace or attach it to their head with tape. The device detects when the wearer is speaking to it. This feature ensures Omi only processes relevant audio input.
Plaud's NotePin is a tiny wearable voice recorder priced at $159. It is one of the pricier options on this list. However, its built-in AI transcription and summarization features make it valuable for lawyers and students. The device records on the wrist or attaches magnetically to clothing.
Recordings are saved in real time on the user's phone. This eliminates the hassle of manual note-taking during meetings or lectures. The device includes 300 free monthly transcription minutes.
Users can upgrade to the Pro plan for $8.33 per month. This upgrade increases transcription time to 1,200 minutes per month. Plaud is preparing to launch a $179 ultra-thin note-taking device called the Note Pro. This new device is now available for preorder.
Rabbit R1 and Sandbar - Retro-style gadgets and the Stream ring features.
The Rabbit R1 is a small, retro-styled handheld device priced at $199. It features a touchscreen and a rotating camera. The device is designed to be phone-adjacent for quick tasks. Users can book flights, order meals, and control apps without pulling out their phone.
Following a crucial software update, the Rabbit R1 now boasts expanded AI features. It introduced "Creations," a feature that allows users to build their own tools and games. This update rectified previous performance issues reported during the initial launch.
Sandbar emerged from stealth last year to unveil the Stream ring. It is an AI wearable designed specifically for note-taking. The ring features a built-in microphone that remains off by default. Users activate it through a flat, touch-sensitive surface on the top.
To capture notes, users press and hold the panel on the ring. They can also interact with an AI assistant through a companion smartphone app. The ring manages media playback, including playing, pausing, skipping tracks, and adjusting volume.
The Stream ring is expected to launch this summer at $299. Silver and gold finishes are available for the new hardware. This price point targets professionals who need hands-free note-taking capabilities.
Why it matters - Implications for model evaluation, privacy, and agent design.
These devices raise complex questions about model evaluation and data privacy. Engineers must test how well models handle continuous listening tasks. Managers need to assess the cost of running models on edge devices versus the cloud.
Privacy is a major concern because these devices record audio constantly. Personal data might be stored locally or sent to remote servers. Regulations like GDPR could impact how companies deploy these products globally.
Agent design is shifting toward persistent memory and context retention. Traditional agents often lose context after a session ends. New agents must maintain long-term memory of user preferences and conversations.
Evaluation metrics must now include accuracy over time and privacy compliance. Standard benchmarks do not cover continuous audio processing well. Teams need new ways to measure performance on these specific tasks.
Safety becomes critical when devices listen to sensitive personal information. Prompt injection attacks could manipulate these devices to reveal data. Input guards must scan prompts before they leave the device.
How OpenSmartRoute helps - Routing audio requests through a secure, scalable system.
A team routing audio requests through OpenSmartRoute gains control over model selection and cost. The router sends each request to the best-fit model, agent, tool, or skill from a defined catalogue. It scores every candidate on quality, cost, speed, and safety.
The team sets weights per request to prioritize specific factors. Hard rules pin a request, ensuring text with personal data stays on an on-premises model. A region or cost cap is never crossed during routing.
The system learns from outcomes so a model that answers well gets more traffic. A model that fails gets less traffic automatically. A new model is simply one catalogue entry that competes on the next request.
OpenSmartRoute works with any OpenAI-compatible provider and open-weight models served locally. It supports MCP tools and A2A agents for complex workflows. A savings ledger shows what each routed request cost next to the most expensive model cost.
An input guard spots prompt injection and personal data before a request leaves the system. The hosted platform keeps a models catalogue with prices and public rankings built from real traffic. This transparency helps managers make informed decisions about hardware and software investments.
What to do - Steps for engineers and managers to test and adopt these tools.
Engineers should test how these devices handle continuous audio and context retention. They can compare the performance of different models on long conversations. Managers should evaluate the total cost of ownership for each wearable option.
Read the privacy policies of each device to understand data handling practices. Check if the device supports local processing or requires constant cloud connectivity. Compare subscription costs against the value of the insights generated.
Consider integrating these devices into existing workflows for note-taking and summarization. Start with a pilot program to measure accuracy and user satisfaction. Monitor the cost of subscriptions and hardware maintenance over time.
Stay updated on new features and product updates from the manufacturers. Watch for changes in pricing or capabilities that might affect your strategy. Keep an eye on regulatory changes that could impact deployment.
AI21 Labs launched LIFT-Bench to fix reward hacking in reinforcement learning training. The new benchmark scores instructions by section to prevent models from cheating.
