The first hardware collaboration in Feishu's history, a "AI Recording Bean" was developed with Anker Innovations.

According to "Intelligent Emergence", Feishu will jointly release an intelligent recording hardware named "AI Recording Bean" with Anker Innovations.

Based on the information we've obtained, this is a round device weighing only 10 grams, with a bean - shaped appearance. The device is equipped with a dual MEMS microphone array, and the total weight of the device with the charging case is about 48 grams. The product supports both Bluetooth and Wi - Fi transmission modes, and features seamless wearing and on - the - go recording.

Since the overseas startup Plaud opened up the market with its AI recording card product in 2025, many domestic manufacturers have followed suit and launched similar products. However, it's rare for large - scale companies to directly enter the market for AI recording hardware.

Since its establishment in 2017, Feishu has never launched any hardware products. Multiple sources close to Feishu told 36Kr that this AI Recording Bean is an exploratory project within Feishu, and the Feishu product team is responsible for the software development.

Since 2024, many innovative forms of AI hardware have emerged in the market, such as recording cards, recording beans, AI portable cameras, and AI glasses. These products all aim to find an AI interaction entrance closer to users' daily scenarios beyond mobile phones.

Instead of following the mainstream card form in the market, Feishu and Anker Innovations chose a more compact "bean" - shaped design. This is a deliberate choice for differentiation.

The advantage of the card - shaped design is that it can be attached to the back of a mobile phone using magnets, but the drawback is that it's larger in size and has more limited wearing scenarios. The bean - shaped design is more similar to the popular AI companion hardware in recent years, emphasizing seamless wearing and all - day portability.

For example, users can clip it on their collars or the edges of their pockets, or fix it on the back of a mobile phone using magnetic accessories. This design reduces the formality of recording, allowing users to start recording at any time during daily meetings, client visits, or even during their commute, without having to take out their phones and open an app specifically.

Feishu x Anker Innovations AI Recording Bean

Another notable design is that the AI Recording Bean supports the real - time meeting minutes function. It's not easy to achieve this considering its size. Users can view subtitles and AI summaries in real - time during recording, and it supports various legend styles such as bar charts.

This design enables the device to obtain more context more quickly and flexibly. According to "Intelligent Emergence", the AI Recording Bean with its charging case can achieve a total battery life of over 32 hours. It has a storage space of 8GB, which can store about 250 hours of audio, and it supports fast charging - 10 minutes of charging can provide 2 hours of recording time. These parameters are at the mainstream level among similar products.

Competing for a closer context entrance

Since 2025, with the further improvement of basic models, especially multi - modal capabilities, the voice interaction field has witnessed a new wave of startup enthusiasm.

In China, established players such as iFlytek and Sogou have upgraded their AI voice input methods. Doubao under ByteDance has also launched the Doubao Input Method, integrating large - model capabilities into daily input scenarios. Additionally, AI application products like Remio also aim to allow users to seamlessly manage the context of their work or life scenarios.

Whether it's AI voice input methods or AI hardware, these startup trends all point in the same direction: competing to become the memory entrance for users' AI assistants. The one who can obtain more user context data can provide more personalized AI services and thus build deeper product barriers.

"Intelligent Emergence" has learned that the AI Recording Bean will cover the entire process from in - meeting to post - meeting. During a meeting, users can see the key points summarized by AI in real - time. After the meeting, the system can quickly generate a complete meeting minutes including speaker roles, to - do items, and key decisions.

Moreover, all recorded content will be automatically stored in the Feishu knowledge base. AI will automatically complete the conversion from voice to text and from text to structured information. Users can retrieve, question, and create secondary content from historical records through the AI assistant in a conversational manner.

For example, users can directly ask the AI, "What concerns did the client mention in last week's meeting?" The system will extract relevant information from the recording and provide an answer. This full - process coverage will significantly reduce users' post - processing costs.

Feishu's product system originated from ByteDance, an organization with high knowledge density. Launched formally in 2017, Feishu chose a differentiated path at the beginning. It entered the high - tech Internet industry and did a great job in functions such as document management, meetings, and transcription, earning its first wave of product reputation.

This strategy has been effective, allowing Feishu to also have a good reputation and product experience in the AI era. For instance, Feishu has a high penetration rate among AI Native companies. Leading large - model companies such as DeepSeek, MiniMax, Dark Side of the Moon, and Zhipu AI all chose Feishu as their internal collaboration tool in the early days of their startups.

The high - tech Internet industry is precisely a knowledge - intensive industry with frequent meetings and numerous documents, which highly overlaps with the target users of AI recording hardware.

From this perspective, Feishu's entry into the AI hardware market is an expected and reasonable step, enabling better collaboration with its flagship products such as Feishu Docs and Feishu Spreadsheets.

The usage habits and pain points of these customers directly determine the product logic of the Recording Bean. For example, meeting minutes need to distinguish speakers, to - do items need to be automatically extracted, and historical records need to be searchable. These are the most common requirements of Feishu when serving such customers.

Regarding the AI recording device market, there is a current consensus: similar recording devices can be replicated at the hardware level in a relatively short time, but the gap in product experience mainly comes from continuous refinement at the engineering level.

This includes insights on different office scenarios, such as continuous optimization of meeting minutes templates and fine - tuning of various details in the post - processing flow. These engineering capabilities require long - term accumulation.

In this regard, Feishu also has certain advantages. For example, the AI Recording Bean is powered by ByteDance's self - developed Doubao large - model, which is at the leading position in multi - modal understanding capabilities. The accuracy of recognition and the ability to understand complex conversations directly affect the accuracy of transcription and the quality of intelligent meeting minutes generation. This is a technological barrier that startups are difficult to catch up with in the short term.