Most AI agents cannot understand the physical world, so we built a POI data layer
Hey Indie Hackers,
We just launched xMap POI Data, a structured point-of-interest data product for developers, AI builders, and location intelligence teams.
The idea came from a simple problem we kept seeing:
AI agents are becoming great at reasoning over documents, code, and text. But when it comes to the physical world, they still need reliable data.
For example, if an agent needs to answer:
- “Find coffee shops within 500 meters of this address”
- “Which neighborhoods have high restaurant density but low direct competition?”
- “Show all gas stations near these highway exits”
- “Find chain stores in this city by category”
- “Build a local search or recommendation tool”
…it needs structured POI data.
Not screenshots.
Not scraped lists.
Not manually cleaned CSVs.
Not one-off lookups.
It needs a proper data layer.
So we built xMap POI Data.
It lets you query or download real-world place data with fields such as business name, category, address, lat/lon, phone, website, hours, rating, price tier, chain, ZIP, city, state, verification status, and more. xMap’s homepage currently lists 24M+ US locations, 30+ fields per record, API/MCP access, and dataset downloads.
You can use it for:
- AI agents
- Local search apps
- Retail analytics
- Site selection
- Real estate analysis
- Travel and tourism tools
- Logistics planning
- Competitive intelligence
- Market research products
The product supports real-time API/MCP access as well as downloadable datasets in formats like CSV, JSON, Parquet, and GeoJSON.
We are especially excited about the MCP angle because we believe AI agents will increasingly need direct access to physical-world datasets. xMap’s broader positioning is becoming “data infrastructure for agentic AI,” where agents can query POI, mobility, traffic, parcel, and demographic layers directly.
Would love feedback from the IH community:
- What would you build with a POI API?
- Which fields matter most to you?
- Would you prefer API access, full dataset download, or both?
- What other physical-world datasets should we add next?
Happy to answer questions.
}
This is a strong direction. The interesting part is not just POI data, it is the “physical-world data layer for AI agents” angle. That feels much bigger than maps or local search, because the buyer is not only someone building location apps, it is anyone building agents that need reliable real-world context.
I’d probably make that positioning even sharper: agents already understand text, code, and documents, but they still need structured access to places, movement, retail density, competition, parcels, demographics, and real-world signals. That makes this feel closer to infrastructure than a dataset product.
The naming is the one thing I’d watch early. xMap works well for map/location use cases, but if the product expands into broader agentic AI data infrastructure, the name may keep pulling people back into “maps” instead of “real-world intelligence layer.” A name like Exirra .com would carry the broader data/AI infrastructure direction better if this becomes more than POI.