Spatial Intelligence,
Verified.

The open-source engine that replaces LLM guessing with ground-truth spatial computation. Topology. Distance. Containment. All correct.

Geodesic Distance ● Verified
—
WGS84 ellipsoid · pyproj Geod.inv · ...
System ...
Models ...
Live ...
Version ...

LLMs guess. GeoSpark computes.

Every spatial answer is computed on the WGS84 ellipsoid, not inferred from training data.

❌ LLM Alone
~500 km
"How far is the Eiffel Tower from Big Ben?"
Guessed from training data. 0% accuracy on distance benchmarks across 4 of 5 model families.
✅ With GeoSpark
343,923 m
"How far is the Eiffel Tower from Big Ben?"
Geodesic computation on WGS84 ellipsoid. Ground truth. Reproducible. Citeable.
0%
Distance accuracy
LLMs alone (4 of 5 families)
535
Benchmark questions
5 categories, 5 model families
100%
Spatial reasoning accuracy
with GeoSpark augmentation

The Platform

Spatial reasoning, live data, autonomous agents, and model-agnostic integration.

🌎

Spatial Engine

Topology, geodesic distance, CRS transforms, buffering, area. Geometrically correct on WGS84.

🧠

Multi-Agent Coordinator

One input, intelligent routing. The coordinator classifies your goal and dispatches to GeoAgent, SpatialReport, or SiteSelector automatically.

🤖

Autonomous Specialists

GeoAgent, SpatialReport (with weather + air quality), SiteSelector. Give a goal, get analysis.

🧠

Spatial Memory

Dual memory: timeless facts + timestamped episodes. Vector-based recall, automatic contradiction detection, auto-linking of related knowledge.

📚

Context Database

Tiered L0/L1/L2 storage for missions, datasets, and analysis history. Hotness scoring and hierarchical retrieval keep prompts lean.

🌤

Live Data Channels

Weather, air quality, fire detection. Real-time from Open-Meteo, OpenAQ, NASA FIRMS.

🔌

MCP + Integrations

6 MCP tools for Claude Desktop. OpenAI, Anthropic, Ollama adapters. Any LLM gains spatial reasoning.

📊

Benchmarked

535 questions, 5 benchmarks, 5 model families. Reproducible and citeable results.

💰

Zero-Cost

9 local Ollama models, free APIs, no rate limits. Run the entire stack on your hardware at $0.

Get Started

# Install
pip install geospark-ai
from geospark import Engine

engine = Engine(tools=["geocoder", "terrain"])
result = engine.ask("How far is the Eiffel Tower from Big Ben?")
print(result.spatial_context.summary)
Open Playground    View on GitHub →