Geocluster Introduces Open-Source Workspace Built for Agent-Driven Geoscience Development

Singapore-based research company, Eigenform, is building environments for the locally operated AI models it believes every business will eventually own.

SINGAPORE, SINGAPORE, October 19, 2026 /EINPresswire.com/ -- What happens to the software industry when artificial intelligence can build the software itself?

Singapore-based AI research and engineering company Eigenform believes the answer is a fundamental shift in what businesses buy and how organisations develop their own technological capabilities.The company anticipates a future in which every business operates its own specialised AI: a persistent system that understands its data, performs its work, and becomes more capable through experience.

Such systems will need more than a chatbot interface. They will require environments in which they can access information, execute tools, evaluate results, and develop new capabilities. Geocluster is designed as one such environment.

Today, the company is launching Geocluster, an open-source, AI-native earth sciences analysis platform built around that conviction.

For decades, specialist software has been sold as a collection of predefined capabilities. Developers build features, customers purchase licences, and more sophisticated functionality typically comes at a higher price.

But as AI coding agents become increasingly capable of developing customised applications, that model is coming under pressure. Geocluster is the first step in a broader effort to rethink how businesses use artificial intelligence.

Today, most organisations access AI through general-purpose models that remain largely unchanged by the work they perform. Eigenform believes the next generation of business AI will look very different: specialised models operating on local infrastructure, adapting to their users, and improving through experience. Under this paradigm, enterprises will be the first to move toward a form of recursive self-improvement, where an AI agent’s own research discoveries become the material it trains on next. Rather than a static tool, Geocluster is built to let agents compound their capabilities over time, each task sharpening the next.

Rather than building another application around today's AI models, Eigenform is developing the environments in which tomorrow's self-improving models will operate.

Geocluster functions as a pre-adapted harness capable of evolving with models - acting as the sensory body and operational ecosystem for local AI agents. Out of the box, Geocluster equips agents with:

- Native Spatial & Geoscience Visualization: Integrated 2D mapping with satellite overlays, shapefile handling, and 3D voxel/sample renderers.
- Automated Format Ingestion: Native parsers for specialized geological file types, table formats, and spatial rasters.
- Execution Safety Rails: Pre-configured environments that allow agents to execute spatial transforms, data cleaning, and dataset profiling without breaking underlying workflows.

Rather than selling a fixed collection of analysis tools, the specialist earth science platform provides AI agents with an environment designed for working with geological data. It combines specialist file handling, analytical tools, scientific visualisations, and an extensible workspace in which agents can develop new functionality in response to users' requirements.

The software is already available as an open source download, with no additional charges for unlocking features. Users can access a free hosted version and purchase additional inference and storage capacity as required, or deploy the software on their own infrastructure.

The company's longer-term ambitions extend beyond earth science and geology, providing local environments and models for new materials, biomedical research, engineering and other research-intensive fields in which proprietary IP is too valuable to allow into the training data of shared commercial models.
Today, most businesses access general-purpose AI models through external services. While those systems can be adapted with company data and instructions, the underlying models do not ordinarily improve their learned capabilities through routine use.

Eigenform CTO Elijah Ng said, “Geocluster was designed to be a review/audit surface first, agent interaction second, and editor last. An inversion of previous generation agent embedded editors. The scientific toolkit built into the software also mirrors the environment that we use when training our agents.”
“I'm a fan of an analogy from XAI dev Poteto: orchestrating agents is like a Michelin kitchen rather than a factory. Yes, you may not have prepared everything, but as the chef, you get the final taste and say over what gets served.”

Eigenform envisions a different future: businesses operating their own persistent AI models, trained on their specialised work, improving over time, and running on infrastructure they control.

Such models will need environments in which they can access data, use tools, execute experiments, and evaluate their own results.
Geocluster is designed to provide that environment for mineral discovery and the resources sector more generally.

CEO Dr. Jen Dodgson said, “Our bet is that future software will be something you build, not something you buy. Imagine two companies. One is looking at floodplain drainage, another is exploring for copper in the desert. Up to now they would have used a lot of the same static software packages. We’re offering something that will adapt around their use cases, learning with them to become an expert in their specific field.”

About Eigenform

Eigenform is a Singapore-based AI research and engineering company working towards self-directed superintelligence: systems that identify problems, form hypotheses, test them against the world, and learn from what they discover. Its work spans recursive self-improvement, agentic AI, inference-time adaptation and AI for scientific discovery, and it builds domain-specific AI platforms for enterprise and scientific clients, with particular experience in geoscience and natural resources.

Akira Rafhael
Eigenform Pte Ltd
ops@eigenform.ai

Building Recursive Self Improvement