Microsoft "Project Solara": The vision of agent-centric hardware

Microsoft's Project Solara aims to establish a new device category with AI agents instead of apps. The focus is on the business sector.

listen Print view
Photo of the reference design for Project Solara

(Image: Microsoft)

6 min. read
Contents

Microsoft unveiled a new software and hardware platform called “Project Solara” at its Build 2026 developer conference. The platform is intended to establish a new device category where AI agents form the central interaction layer instead of classic apps.

As Steven Bathiche, CVP & Technical Fellow of the Applied Sciences Group at Microsoft, explained, a key feature of Project Solara is the separation of devices and intelligence. Unlike Microsoft's Surface RTX Spark Dev Box, announced at the same time and intended to run local AI workloads with significant computing power, no models are locally intended on the Solara devices. The devices are intended to function as slim terminals – as windows into a cloud-based agent infrastructure based on Azure.

The devices are to be based on MDEP (Microsoft Device Ecosystem Platform) – remarkably not Windows, but an operating system based on the Android Open Source Project (AOSP). Microsoft already uses MDEP for Teams Room devices; for the power-efficient, resource-saving hardware of the Solara devices, Android is likely the more pragmatic choice than Windows. Management is intended via Microsoft Intune, authentication via Entra ID and Hello for Business with biometric methods (fingerprint or facial recognition). Physical privacy controls such as a hardware microphone mute switch are planned.

The agent interaction model is intended to work with so-called “Just-in-Time UI”: the devices are not designed for classic apps, but for an adaptive UI layer that is intended to dynamically adapt to the device, screen size, and interaction mode – visually, by voice, or multimodally. Microsoft describes this as a spectrum between classic responsive design and fully generative UI. The latter is still a pipe dream.

Project Solara is intended to be designed as an open multi-agent system. Companies should be able to integrate their own agents via Copilot Studio, the Microsoft 365 Agents SDK, or the Azure Agent Framework. A single-agent lock-in is explicitly not the goal.

Videos by heise

Microsoft has won Qualcomm and MediaTek as hardware partners. Both supply commercially available chips – which, according to Microsoft, is central to being able to realize new form factors cheaply and quickly.

The portable badge device is based on a Qualcomm chip and adopts the form factor of traditional access cards – a device that millions of employees carry daily. A touchscreen display, a side-mounted fingerprint sensor, a camera, a microphone array with speaker, and 5G, WLAN, Bluetooth, and GNSS connectivity are planned. This does not fit into a classic smart card; the size of the badge is more like a CI module.

Steven Bathiche showed the reference design of an intelligent access card during his presentation of "Project Solara", which can be used to control agents.

(Image: Microsoft)

The stationary device is reminiscent of Amazon's Echo Show devices in terms of form factor. It is based on a MediaTek chip and is designed as a desktop companion – not a PC replacement, but a permanently available agent access point. A touchscreen display, facial recognition, a UWB presence sensor, a dual microphone array, and two USB-C ports are planned. Via USB-C and an external monitor, the device can be used as a Windows 365 client. It can be paired with a Windows PC via Bluetooth to transfer tasks between devices.

The stationary Solara device is reminiscent of Amazon's Echo Show.

(Image: Microsoft)

Microsoft names the clinical area as one of the core application cases. Clinic staff already wear ID cards today – but so far exclusively as passive RFID or magnetic cards for access control and medication dispensing. The Solara badge is intended to expand this function with an active agent layer: With the patient's consent, the device is intended to record, transcribe, and diary conversations – i.e., automatically divide speakers. An agent is then intended to store the structured data in the patient record. Microsoft names Dragon Copilot as the intended solution for this.

In this context, the cloud architecture is less of a limitation than a requirement: clinical language models would be too large for a wearable, patient data must be stored centrally and auditable, and model updates can be deployed centrally without having to update thousands of devices individually.

Within Microsoft, hundreds of employees are already testing the concept devices in their daily work. An external pilot program with corporate partners is planned in the coming months, including AccuWeather, Best Buy, CVS Health, Levi's, and Target. In the medium term, OEMs are intended to develop their own solutions based on the reference designs for industries such as healthcare, retail, hospitality, financial services, and manufacturing.

The idea of replacing apps with “agents” as the primary interaction layer is not new: Rabbit R1 and Humane AI Pin have already attempted this paradigm shift – and failed, albeit in the consumer market and with immature technology. Project Solara takes a structurally different approach: B2B instead of B2C, specialized workflows instead of all-purpose devices, embedded enterprise infrastructure instead of proprietary ecosystems.

Whether agent technology is now mature enough to fulfill the promises of reliable context sensitivity in everyday work will have to be shown by the announced pilot programs. Microsoft itself phrases it cautiously: “We are still early.”

(vza)

Don't miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.