GeoQ AI Agent

Ask the Map Anything —
AI Handles the SQL

GeoQ's AI agent translates plain-English questions into optimised SQL, executes them against your registered geospatial sources, and returns summarised results with automatic map rendering.

Try AI Chat → See Full Web Platform
How can GeoQ's AI agent help?
Ask in plain English — get SQL, GeoJSON, and map results.
GeoQ AI Chat Live
Show me all buildings taller than 50 m within 500 m of the river in the current map view.
Found 34 buildings matching your query. Rendering on the map now.
SELECT name, height, geom FROM features WHERE height > 50 AND ST_DWithin(geom, river.geom, 500)
✓ 34 features · GeoJSON · 0.28s
Now filter to only those built after 2010 and colour them by height.
Refined to 18 buildings. Applied graduated colour ramp (yellow → red) by height. Tallest: Riverside Tower at 112 m.
... AND year_built > 2010 ORDER BY height DESC
✓ 18 features · map updated
Under the Hood

NL → SQL → GeoJSON in One Round-Trip

Your Question
Plain English
Schema Context
Table + column introspection
FlockMTL / OpenAI
llm_complete → SQL
DuckDB Execute
query_arrow()
GeoJSON + Summary
Map update + prose
AI Capabilities

More Than Just SQL Generation

Schema-Aware Prompting
The agent introspects your registered sources — column names, geometry types, bounding boxes — and injects that context into every prompt for accurate SQL generation.
Semantic Filtering
Use llm_filter to semantically filter rows that traditional WHERE clauses can't reach — e.g. "features that look residential" from a free-text description column.
Result Summarisation
After query execution, llm_reduce compresses thousands of rows into a concise analyst-friendly answer — counts, extremes, patterns, and anomalies called out automatically.
Semantic Embeddings
llm_embedding powers vector similarity search over metadata — find datasets "similar to flood risk maps" from a catalog with hundreds of entries.
Relevance Reranking
llm_rerank orders multi-source results by semantic relevance to your question — putting the most important features at the top of the map and result table.
Local & Private LLMs
FlockMTL supports Ollama for fully local inference — no data leaves your environment. Switch between OpenAI, Anthropic, and Ollama via a single CREATE SECRET statement.
REST API

Chat Endpoint

// POST /api/chat { "message": "Which census tracts have population density above 10,000/km²?", "bbox": [-122.5, 37.7, -122.3, 37.9], "session_id": "abc123" } // Response { "summary": "Found 8 census tracts with density above 10,000/km². Densest: Mission District at 18,420/km².", "sql": "SELECT geoid, pop_density, geom FROM features WHERE pop_density > 10000", "geojson": { "type": "FeatureCollection", "features": [ ... ] }, "result_type": "geojson", "row_count": 8, "session_id": "abc123" }

Let AI Do the SQL

Set OPENAI_API_KEY, start geoq-web, and start chatting with your data.

Buy GeoQ → ← Back to Home