At InfoComm 2026, video collaboration vendor Neat demoed two beta products built on a single architectural bet: that meeting room AI should run on the device and on the customer's network, not as a cloud inference call.
Your meeting room is supposed to be ready when you walk in. It almost never is. The cable stops working, the camera frames the back of someone's head, the lobby room never switched back from the last call. The fix is usually an IT ticket that takes hours.
That operational pain is the anchor Neat, a video-collaboration hardware vendor from Oslo, used this week at InfoComm 2026 in Las Vegas to position two new on-device AI capabilities, one aimed at the people who run meeting rooms and one at the people in them.
InfoComm is the pro-AV industry's annual trade show, the place where video-conference room vendors demo what is about to ship to enterprises. Neat's booth this year made two announcements: Neat Pulse MCP and Intelligent Framing. Both are built on the same architectural bet, that meeting-room AI should run on the device and on the customer's network, not as another cloud inference call.
The bigger of the two is Neat Pulse MCP. It is a bridge between Neat's existing Pulse fleet-management console and large language models, including Anthropic's Claude and Google's Gemini CLI. Instead of logging into a dashboard and clicking through device groups, an IT admin types a request, in something close to plain English, into a CLI client, and an AI agent handles the routing. "Intelligent agents can now manage devices across environments," said Tormod Ree, Neat's senior vice president for engineering, framing MCP as the part that makes that workflow practical.
The acronym is the interesting bit. MCP stands for Model Context Protocol, an open standard originally popularized by Anthropic for letting large language models talk to external tools and data sources. Enterprise IT has started paying attention to it as a way to give AI agents real operational reach without sending every command and every log line to a third-party model. Neat's bet is that the meeting-room fleet is a natural place for that pattern to land: dozens or hundreds of devices per site, repetitive workflows, clear policy boundaries, and a customer base that already cares about where its data goes.
That is also why Pulse MCP is being shipped, for now, as a locally-hosted beta. It runs against the customer's own LLM, not against Neat's or any model provider's hosted endpoint. The positioning is direct: device data, meeting recordings, and operational telemetry stay on the customer's side of the firewall.
The second piece is Intelligent Framing, an on-device AI that decides who the camera should show in a large meeting. It runs three modes: Equal, a Highlights view that dynamically tiles three, six, or eight active speakers, and a Highlights + Room option that keeps a defined number of close-ups (three, four, or five) next to a wide shot of the room. The processing happens on the Neat hardware, not in a cloud inference call, the same architectural story Pulse MCP is telling for IT: keep the AI close to the device, keep customer data on site.
That parallel is what makes the announcement worth reading as more than two product launches. Neat is positioning itself as "AI-native" hardware and software for workplace collaboration, and both products are framed around distributed, edge-resident AI instead of rented cloud inference. It is a meaningful contrast with how most meeting-room AI has shipped so far, which has leaned on cloud transcription and cloud summarization services. Whether that contrast turns out to be a category shift or a marketing re-label of work that was already happening with prior auto-framing and remote-management tools is the question neither product can answer by itself.
The honest limit, and the one IT teams should keep in mind, is that neither product is generally available yet. Neat Pulse MCP is in a locally-hosted beta for existing Pulse customers. Intelligent Framing is in beta, with general availability targeted for the second half of 2026. Both were shown as live demonstrations at InfoComm, not as deployments in named customer environments. "Agentic AI is reshaping how hybrid teams collaborate, and meeting rooms are an obvious place for that to play out," said Javed Khan, Neat's chief executive, in the announcement, which is on-the-record vendor commentary, not independent validation.
What would actually move the story from announcement to trend: a third-party integrator or enterprise running Neat Pulse MCP against a real fleet of devices, with metrics on time-to-resolution or ticket volume. So far the source cluster for the announcement is the vendor itself, through the press release picked up by ITBusinessNet and the two parallel Neat blog posts. Independent analyst reads on whether MCP-based fleet agents and on-device meeting AI are an actual category shift are not yet in the record.
For now, the bet is simple and testable. Neat is placing its next platform story on an architectural choice, local hosting and agent-style operation through MCP, that enterprise IT teams have been asking for in adjacent categories. Whether the architecture survives contact with a real fleet, a real budget cycle, and a real security review is the next question, and the one the press release cannot answer.