OpenSpace used its Waypoint 2026 virtual customer event on 10 September to announce the next generation of its Visual Intelligence Platform, framing the release as a move beyond document-centric workflows. The argument carrying the announcement is direct: AI agents that only read text cannot verify what is happening on a physical jobsite, and the company intends to make its visual data the grounding layer other agents consume.
What the platform update includes
The announcement spans several capabilities, most of which are in preview or Early Access rather than generally available. AI Autolocation 2.0 with Live Location provides indoor positioning that the company says works like GPS inside buildings without requiring installed hardware such as bluetooth beacons. The 2.0 version adds real-time location tracking.
Site Mode places a user's live location directly on project drawings and BIM models, even offline, letting field teams access information and file Field Notes based on where they are standing. AI Walk-and-Talk lets users create structured updates by speaking as they move through the site, with images and voice tagged to locations in real time.
On the progress-tracking side, OpenSpace Track now includes quantity tracking, markups and predictive analytics. The company also announced a Track API that lets customers pull progress data into their ERP, BI dashboards or project management tools. The OpenSpace Agent Ecosystem was described as an open, interoperable platform where AI agents can access spatial data to understand site conditions and act on them.
The strategy: agents need eyes
Co-founder and CTO Michael Fleischman made the positioning explicit: "AI agents need eyes. An agent may be able to read every report associated with a project, and that provides real value. But without visual and spatial context, AI still cannot fully understand the scope of what drives business results for our customers - that data lies in the field."
That claim targets the document-centric agents the construction software market has shipped through 2026. Most agents released this year parse text - RFIs, submittals, specifications, daily logs - but have no way to verify whether what a document describes matches physical reality on site. OpenSpace's answer is to treat visual capture as the grounding layer and rent access to it, rather than building every agent itself.
CEO and co-founder Jeevan Kalanithi framed the same idea as a refusal to add another data-entry tool: "Construction doesn't need another disconnected tool or system that asks teams to manually input information into a form. The next phase of the industry will be defined by platforms that understand the physical world and make that intelligence available wherever teams already work."
The Track API reinforces the posture. Pushing progress data into a customer's existing ERP or BI stack means competing to be a data source rather than the system of record. That is a different bet from the direction taken by Procore, which agreed to acquire DroneDeploy for roughly $845 million in July, or Buildots, which raised $130 million for its own construction intelligence platform.
The data behind the argument
OpenSpace cites four company metrics to support its scale: imagery analysed from more than 110,000 construction projects, over 77 billion square feet captured, across 132 countries, and more than 500 million expert-verified labels and descriptions. The last figure matters most for the agent pitch. Captured square footage measures what cameras have seen; half a billion expert-verified labels is a claim about supervised training data, which is what makes a visual model useful to downstream systems. The release does not define what qualifies a label as expert-verified, and none of the four figures have been independently verified.
On the customer side, the release names Suffolk, Comfort Systems, and Tishman Speyer as companies that rely on OpenSpace for site insights. These appear in the boilerplate as general customers, not as participants in any of the newly announced capabilities.
What is and is not shipping
The company is explicit that much of what was shown is not generally available. The release uses terms like "previewed," "upcoming capabilities" and "in Early Access," and says additional information on feature availability, Early Access and Beta programmes, and general customer access "will be shared in the coming months." No pricing or general availability dates appear in the announcement.
What the release does claim as live is the programme behind the Agent Ecosystem: "OpenSpace is already putting this vision into practice, working with a select group of Early Access customers to build and deploy AI agents against real-world construction workflows." The Early Access group members are not named, and no third-party agent or developer in the ecosystem is identified. The Agent Ecosystem should be read today as a declared direction with an Early Access programme, not as a live marketplace.
Why this matters for operations, real estate and construction professionals
The practical question is whether visual grounding data makes agents more useful enough to justify adding another data layer to the stack. Document-reading agents are already in production and improving. If OpenSpace is right, the gap between what a report says and what a site shows is the next bottleneck - and teams that connect the two will resolve issues faster. If the document-centric approach proves sufficient for most workflows, the market may not need a separate spatial intelligence layer. The answer will emerge as Early Access customers put agents into real workflows, but for now both paths are open.
Your membership also unlocks:
