Startup Ambi, founded by former Plaud executive Peter Mo, has raised $26 million to develop an AI assistant capable of analyzing a user’s conversations and work activity throughout the day. The service is designed not only to record meetings and generate summaries, but also to offer real-time prompts, find relevant information, and help complete tasks directly during conversations.

At the same time, the product raises privacy concerns: continuous context collection affects not only the user, but also the people around them, who may not know their conversations are being recorded.

AI that helps during the conversation itself

Ambi is releasing apps for iPhone, Mac, and Apple Watch. In terms of how they work, they resemble AI services for recording meetings, but users can allow them to collect context throughout the entire day rather than only during individual calls.

The system analyzes calendar data, previous meetings, knowledge bases, and the internet. Based on that information, the AI agent can suggest relevant information or prompt the user on what to do next while the meeting is still in progress.

The developers want to go beyond ordinary notes and task lists that services generate only after a conversation ends. Ambi is intended to help directly in the workflow: finding needed information, suggesting solutions, and launching task execution. Work started on one device can be continued on another.

According to Peter Mo, traditional AI tools are often tied to a screen or require users to first record a conversation, transfer the recording to a phone, upload it to the cloud, and wait for processing. Ambi aims to make assistance available at any moment, without the need to start a separate process each time.

The company is preparing an app that will remember work done on a computer

The startup plans to expand context collection beyond conversations. A future desktop app will be able to analyze user actions on a computer and store important details in memory in order to provide more personalized recommendations.

Other services already use a similar approach, including Littlebird and Minimi. Their goal is to make information about everyday work available for later queries so that users do not have to repeatedly explain circumstances and tasks to the AI from scratch.

Ambi also plans to release its own device with a screen, speaker, camera, and microphones. It is intended to gather information about the surrounding context and display useful information directly on the device.

The company raised $26 million across several investment rounds. Lead investors include Luminous Ventures, which previously invested in Meituan and Unitree Robotics, and Chinese electronics manufacturer Huaqin.

The startup has more than 50 employees in San Francisco, Beijing, Shanghai, and Shenzhen. The company is also looking for business development specialists and engineers working on speech recognition models.

The main risk is conversations involving people who did not give consent

Always-on listening could make the AI assistant more useful, but at the same time it creates serious privacy risks. Unlike a regular voice recorder that is turned on before a specific meeting, background data collection can potentially cover many conversations and situations throughout the day.

Meanwhile, people nearby may not know that their words are being recorded. TechCrunch notes that it is still unclear how visibly Ambi’s apps and future device will warn other people about audio context collection if the user does not disclose it themselves.

The problem is not limited to a single startup. Some developers of AI services for meeting transcription have already faced lawsuits over recording conversations and processing data without participants’ consent. Apple’s new features related to continuous speech transcription have also raised privacy concerns.

Peter Mo says that background data collection is turned off by default in Ambi and that users can stop it at any time. According to the company, it does not store users’ audio recordings and encrypts data, although the startup has not disclosed the specific encryption method.

But even if recordings are not stored, questions remain about how other information extracted from conversations is processed and how long it is retained. It is equally important whether a nearby person will be able to tell that their speech is being analyzed.

Mo acknowledges that the industry has not yet solved the problem of informing bystanders. According to him, the company tries to remind users of the need to notify other people about background context collection, but it cannot guarantee that they will not hide it.

This is exactly where the core contradiction of such assistants arises: the more information they collect, the more useful they can potentially become — and the more people are affected by issues of consent and privacy. Ambi’s success will depend not only on the quality of its AI, but also on how convincingly the company can address these problems.