How Arnis Uses GitHub Copilot and AWS to Scale Open-Source Geospatial Data Processing

The open-source project Arnis, which converts real-world OpenStreetMap data into 3D voxel environments, has reached nearly 900,000 downloads since its inception. To manage this scale, its sole maintainer has integrated GitHub Copilot sub-agents for automated regression testing and migrated the project’s elevation data pipeline to AWS Terrain Tiles. For IT decision-makers and software engineering leads, Arnis provides a practical blueprint for leveraging AI developer tools and public cloud data registries to maintain complex, high-volume software without incurring prohibitive infrastructure costs.

How GitHub Copilot Sub-Agents Manage Arnis Code Regressions

Managing a rapidly growing open-source codebase as a sole maintainer requires strict automation. Louis, the creator of Arnis, reviews every pull request personally, a process that historically created a bottleneck for community contributors. To resolve this, he integrated GitHub Copilot directly into his development workflow to handle initial code reviews, performance profiling, and documentation updates.

The most significant architectural shift in his workflow is the deployment of AI sub-agents for quality assurance. Louis maintains a manual inventory of fragile codebase areas that require strict oversight during modifications. Rather than relying solely on standard continuous integration tests, he deploys specialized AI agents to analyze these specific modules.

"I’m starting a few sub agents with the GitHub Copilot app to have a look specifically at those cases, and to come up with corner cases to test to make sure that we do not introduce any regressions into new releases."

This approach highlights a maturing use case for AI coding assistants. Rather than merely generating boilerplate code, AI agents are being deployed as autonomous reviewers that understand the contextual performance implications of specific code changes. As Louis noted regarding the performance sensitivity of the rendering engine, "It takes just one wrong calculation in street lights, for example, to make the generation time of an entire city much slower." By delegating the detection of these micro-inefficiencies to AI sub-agents, maintainers can focus on high-level architectural decisions.

Why Arnis Migrated Geospatial Elevation Data to AWS Terrain Tiles

The core function of Arnis relies on querying the Overpass API for OpenStreetMap data, parsing nodes, ways, and relations, and converting them into renderable elements. However, generating a flat 3D environment is insufficient for realistic terrain. Accurate elevation data is required to replicate real-world topography, such as the steep gradients of Niagara Falls or the rolling hills of rural landscapes.

Initially, the project relied on a commercial elevation data provider. As the user base expanded, the cost of making millions of tile requests for terrain generation became unsustainable. In 2025, the project executed a major infrastructure pivot, migrating to AWS Terrain Tiles hosted on Amazon S3.

This migration eliminated data retrieval costs for millions of tile requests by leveraging the AWS Registry of Open Data. While the GitHub repository tracks nearly 900,000 total downloads, the AWS public sector blog notes that the platform currently serves nearly 300,000 active users. For businesses building geospatial applications, this transition demonstrates the financial necessity of decoupling data retrieval from proprietary commercial APIs and migrating to open, cloud-hosted datasets hosted on object storage like Amazon S3.

How the Meld Orchestrator Scales Arnis for Continental Map Generation

The base Arnis desktop application processes geographic selections sequentially, which is efficient for neighborhood-sized areas but computationally restrictive for larger regions. To address this limitation, the community developed Meld, an orchestrator that splits an OpenStreetMap selection into tiles.

Meld operates by distributing the rendering workload. It breaks a massive geographic bounding box into smaller, manageable tiles and builds them in parallel using a custom fork of the Arnis engine. Once the parallel processing is complete, Meld merges the individual tiles into a single, seamless environment with shared elevation and seed data. This distributed computing pattern allows the software to support Minecraft 1.21+ environments at a continental scale.

This parallel processing capability is also accessible via MapSmith, the browser-based MapSmith interface that allows users to generate worlds directly from a mobile device or desktop browser without installing local software. MapSmith leverages this backend orchestration to handle large areas up to 500 km² in size, roughly eight times the landmass of Manhattan. The shift from a local desktop execution model to a centralized, parallel-processing web interface represents a standard scaling pattern for heavy computational workloads.

What Arnis and MapSmith Mean for Enterprise Geospatial Data Pipelines

While Arnis is primarily utilized for creating interactive 3D block environments, its underlying data architecture mirrors the workflows of enterprise Geographic Information Systems (GIS). The process of querying the Overpass API, interpreting OSM tags, determining geometry (points, lines, polygons), and applying element processors to place blocks is functionally identical to how enterprise software renders digital twins for urban planning or logistics routing.

For IT leaders and development managers, the Arnis project offers three concrete takeaways for internal tool development:

• AI-Assisted Maintenance of Glue Code: The most complex part of any data pipeline is the "glue code" that translates raw data formats into renderable outputs. Using AI sub-agents to monitor and test these specific translation layers prevents performance degradation without requiring manual code reviews for every minor commit.
• Open Data Registries Reduce Variable Costs: Relying on commercial APIs for foundational data like elevation or mapping introduces variable costs that scale linearly with user growth. Migrating to public cloud open data registries (like AWS Terrain Tiles) converts variable API costs into predictable, low-cost object storage retrieval.
• Distributed Orchestration for Heavy Rendering: When client-side or single-threaded processing hits a ceiling, implementing an orchestrator (like Meld) to tile, parallelize, and merge workloads allows applications to scale from local neighborhoods to entire regions without rewriting the core rendering engine.

The evolution of Arnis from a two-week Python prototype to a highly optimized, AI-maintained, and cloud-integrated platform illustrates how modern developer tooling can extend the lifespan and scalability of open-source infrastructure. By combining AI-driven regression testing with cost-effective cloud data storage, maintainers can support massive global user bases while keeping operational overhead near zero.