A New Startup Wants to Give AI Accurate Information About the Physical World

Mireye, a startup backed by Y Combinator, has launched a service that gives AI programs reliable, sourced information about any location in the United States. You can think of it as a reference library that an AI agent can query whenever it needs to answer a question about a specific place. The product went live via a post on Hacker News (source) and is documented at docs.mireye.ai, with a skills file at mireye.com/skills.md designed to be pasted directly into an AI agent's instructions (source).

The problem Mireye addresses is straightforward. Today's AI language models, when asked specific questions about specific places, often make things up. A Fortune 500 insurance company told Mireye's founder, Ansh, that its engineers had given up on building AI tools for insurance underwriting for exactly this reason. Ansh hit the same wall himself while building AI tools for construction that could read online content but had no knowledge of what was physically true about the ground at a given location (source).

Mireye's solution is not an app you use directly. It is a data service that other software, including AI agents, can call. The company's first product was a niche site-screening app, which Ansh discontinued after discovering that customers were querying 311 of 317 data fields with no single dominant use case. The signal was clear: customers wanted the underlying data, not a specialized tool (source).

That data layer now includes 366 fields of information, drawn from 85 authoritative sources. Most of it is US federal data, from agencies including USGS, FEMA, NOAA, USDA, BLM, EPA, and EIA, with additions such as Overture Maps and Sentinel-2 satellite imagery. Coverage spans the full United States, with drive-time analysis additionally covering Canada (source).

What sets Mireye apart from a simple data dump is how it handles provenance and missing information. Every piece of data comes with four things attached: the source name, the source URL, a timestamp of when it was fetched, and a confidence level. Each field also returns one of three statuses: ok, absent, or failed. The distinction matters. An absent status tells an AI that a flood zone genuinely does not exist at a location. A failed status means the county never mapped floods at all. For an AI making a decision based on this data, that difference is important (source).

Data freshness varies by source. USGS elevation data is effectively permanent. FEMA flood maps update on FEMA's own schedule. USDA cropland data updates yearly. Mireye gathers each source county by county, in whatever format a given county publishes, normalizes everything into one consistent format, and refreshes continuously. Where a needed dataset does not exist in any index, Mireye has filed public records requests to obtain it, as it did for sewer main data in North Carolina (source).

The company also offers on-demand indexing. An automated system researches candidate sources, collects data, tests it against known ground truth, and indexes the requested field, usually within a day. Every plan includes field requests, so users can order data fields that do not yet exist. Mireye researches, builds, and validates the field, then notifies the user when it goes live permanently (source).

Pricing is credit-based with a free tier of 5,000 credits per month, no credit card required. Paid plans start at $19 per month. No plan bills past its allowance; usage simply stops rather than accruing extra charges (source).

Mireye's stated mission is to "index every inch of the earth and make it as queryable as the web" (source). The current catalog covers the US, with limited Canadian coverage for proximity analysis. The company is listed under both Y Combinator's Artificial Intelligence and Infrastructure startup categories (source).

In my view, the design choices here reflect a real understanding of where AI systems actually fail. The combination of per-field source tracking and the distinction between "this does not exist" and "we could not find this" addresses a specific, documented problem: AI models that invent place-specific facts because their training data contains no ground truth for a given location. The on-demand indexing model is the more ambitious bet. If it scales beyond sewer mains in North Carolina, it turns Mireye from a static catalog into something closer to a crawler for the physical world, filling gaps in the long tail of county-level data that no single provider has incentives to organize. Whether that model holds at volume is an open question, but the early customer list spans insurance portfolio screening, warehouse sourcing, data center site selection, drone deployment planning, and signals for human trafficking investigations, which suggests the demand for reliable ground truth is broad and not confined to a single industry.