--- title: "AI on Polaris: MCP Server, Fleet Diagnostics & Roadmap | Mersive" canonical: https://www.mersive.com/platform/ai description: "Where AI fits on Polaris, and how Mersive thinks about it: AI serves the IT manager and the end user, and building for them differs. An MCP server ships Q4 2026, with on-device silicon already in every pod." language: en-US publisher: Mersive Technologies --- # AI that manages the room, not the meeting. Polaris is a fleet of Linux devices reporting to one cloud, which makes it a good place for AI to do unglamorous work. It can find the room that is about to fail, explain why, and fix it. That is where Polaris AI starts. Built for it, not retrofitted ## The room already has the silicon. Polaris was designed from the ground up to run AI in our products rather than to have it added later. Polaris Pro carries an on-die NPU rated at 9.6 TOPS at minimum precision, which is enough to run a language model at the edge, so the data and the processing stay on the device and in the room instead of leaving the building to be inferred somewhere else and coming back. The silicon is only half of it. Every room reports to one cloud over one outbound connection with one identity per device, and that fleet is reachable through a documented API, so what makes Polaris a place to put AI is as much how the cloud and the APIs are built as what is inside the box. We think AI will keep producing real innovation in AV on both sides of the room, for the people in the meeting and for the administrators who run hundreds of them. What’s coming next ### As we approach AI feature releases we will list them here for the next quarter. ### MCP server SHIPS Q4 2026 A Model Context Protocol server for Polaris Cloud, so an AI assistant can query and act on the fleet directly. Ask which rooms are unreachable, why a room rebooted last night, or which sites are behind on firmware, and get an answer from the fleet rather than from a dashboard someone has to read. ### AI on the pod Both shipping chassis already carry the silicon. Polaris Pro runs an on-die NPU, which MediaTek calls the APU. MediaTek’s documentation rates it at 4.8 TOPS (INT8), or 9.6 TOPS at minimum precision (INT4), the convention MediaTek uses for its current Genio series. Polaris Essentials carries 3.2 TOPS. ### In the meeting, accessibility first SHIPS Q4 2026 The first AI most participants will meet is live subtitling. It’s an accessibility feature before it’s an AI feature and just as useful for anyone following a meeting in their second language. How we think about it ## Two audiences. One fleet. Not one problem. AI in a meeting room is worth something to two people who are never in the room at the same time. One is the IT manager who runs a thousand rooms; the other is the end user sitting in one of them. For the **IT manager**, this is a retrieval problem. Every Polaris room already reports to one cloud, over one outbound connection, with one identity per device, so an assistant that can answer “which rooms are about to fail and why” is reading data that exists today. That is why the first step on this side is an [MCP server](https://www.mersive.com/platform/cloud). The hard part was never the model; it was having the fleet in one place to ask about. For the **end user**, it is a real-time problem, and that is a different engineering exercise entirely. It happens inside a live meeting, where latency, where audio is handled and where inference actually runs are the design rather than details to settle later. IT manager · MCP server SHIPS Q4 2026End user · live subtitling SHIPS Q4 2026 Polaris + AI ### Operate your entire fleet. In plain language. Ask. Diagnose. Act. Polaris turns natural-language intent into governed operations across every room. 1. 2. 3. 4. Your AI agentConnected through Polaris Connected YouWhich rooms dropped shares this week? Polaris agent Three rooms need attention. Root causes, correlated across the fleet. intentresponse Model Context Protocol Polaris MCP Server Makes governed Polaris capabilities discoverable to compatible agents. Ships Q4 2026 analytics.query devices.reboot firmware.stage analytics.export signage.publish analytics.querystructured result Control plane Polaris Cloud The same secure API your dashboards and ITSM workflows use. Identity & authorization Fleet context Policy enforcement Every operation is auditable governed actionlive telemetry Managed fleet Every room, one control plane 47 SPACES · 2 SITES Headquarters5 rooms online 2.01 2.06 3.02 4.12 4.15 Campus West · Classrooms42 rooms online C-01 C-07 C-14 C-19 C-22 **3 signals correlated** across health, firmware and network telemetry. **Workspace content stays outside the loop.** The agent operates on authorized fleet metadata and controls. **No bespoke integration layer.** MCP describes Polaris tools to compatible agents at runtime. The part that is already true ## The fleet is the dataset. The NPU silicon at the top of this page covers the room. The cloud covers the fleet. Every room already reports to one cloud, over one outbound connection, with one identity per device, and that fleet is reachable through a documented API rather than a bespoke integration. Polaris is a good place to put AI as much for how the cloud and the APIs around the box are built as for what is inside the box, and the fleet data an assistant would need already exists in the product today. [What Polaris Cloud runs today →](https://www.mersive.com/platform/cloud) [Polaris Cloud as a product →](https://www.mersive.com/cloud) [How the device is secured →](https://www.mersive.com/security)