Using the AI Agent
Ask the AI agent to research markets, find properties, and analyze sites.
Using the AI Agent#
Pillar's AI agent can search parcels, find properties, pull GIS data, research markets on the web, and chain these tasks together — all from a conversational interface on the map.
Opening the agent#
There are two ways to start a conversation:
- AI button — click the floating AI button in the bottom-right corner of the map
- Universal Search — type a query in the search bar and click the Ask AI row at the top of the results
The agent panel opens as a dark sidebar on the right side of the map.
What you can ask#
The agent combines three specialist capabilities:
Property search#
Ask the agent to find parcels matching specific criteria. It can search across 160M+ US parcels by location, acreage, zoning, owner type, use code, and more.
- "Find vacant land over 10 acres in Fulton County, GA"
- "Industrial parcels near rail lines in Houston"
- "Waterfront lots under $500k in Miami-Dade"
GIS and data layers#
The agent can discover and use GIS data layers as spatial filters — infrastructure, environmental boundaries, school districts, and more. It can also pull data from external ArcGIS servers.
- "Show me parcels within 2 miles of a substation in Travis County"
- "Find land outside the floodplain near I-85"
Web research#
The agent can search the web, read pages, and pull information from public sources — news, SEC filings, company sites, and more.
- "What's the recent development activity in downtown Nashville?"
- "Research the owner of this parcel"
Multi-step workflows#
The agent can chain these capabilities together. For example, it might download a GIS layer of transit routes, then use that geometry to find parcels within walking distance, then research the market conditions in that area.
How it works#
When you send a message, the agent evaluates whether your query is specific enough to execute immediately or needs clarification first.
Direct execution — if your query includes a clear location and concrete filters (e.g., "10+ acre parcels in Douglas County, CO"), the agent runs the search immediately.
Plan mode — if your query is broad or ambiguous (e.g., "datacenter sites near Atlanta"), the agent first presents a structured plan showing the proposed search area, filters, and spatial constraints. You can approve the plan or refine it before the search runs.
Working with results#
When the agent finds properties, a result card appears in the chat showing:
- Dataset name and feature count
- The search filters that were applied
- An eye toggle to show/hide the results on the map
From the result card, you can:
- View on map — toggle the results layer on to see parcels highlighted
- Edit style — customize how the results are displayed (colors, labels)
- Save to my layers — persist the results as a layer in your Layer Library
Results are temporary unless saved. Saving stores them as a permanent layer accessible from the Layers panel.
Chat features#
- Follow-up questions — continue the conversation to refine results or ask related questions
- Queue messages — type a follow-up while the agent is still working; it will be sent automatically when the current task completes
- Stop — click the stop button to cancel a running task
- Session history — access previous conversations from the menu (three dots in the header). Pin important chats for quick access.
- Multiple sessions — switch between active conversations using the tab bar
Suggestions to get started#
If you're not sure what to ask, the agent shows three starter prompts:
- "Datacenter sites near Atlanta"
- "Waterfront lots under $500k in Miami"
- "Industrial land near rail lines"
Click any of these to kick off a search, or type your own query.
Tips#
- Be specific about location — named counties, cities, or addresses work best
- Include quantitative filters when possible (acreage, price, distance)
- The agent works best for site selection queries — finding land, analyzing locations, and researching markets
- If the agent gets stuck, try rephrasing your question or breaking it into smaller steps