Today, Kuartal introduces Leo-Regulus 1.0, the first generation of its sovereign artificial intelligence model family, designed to bring specialised, locally operated AI to Indonesian governments, financial institutions, businesses, and other organisations.

Leo-Regulus 1.0 marks the beginning of Kuartal's approach to developing artificial intelligence: rather than building one general-purpose model to perform every task, Kuartal is developing a family of specialised AI models that work together, understand their respective industries, and can be adapted to each organisation's existing systems.

Built upon capable open-weight foundation models and developed through additional fine-tuning, Leo-Regulus prioritises three capabilities: understanding Bahasa Indonesia, working with documents and institutional knowledge, and running privately within an organisation's own infrastructure.

The result is a foundation for AI systems that can read, write, retrieve information, and assist with everyday institutional work without depending on external cloud inference services.

Intelligence should belong to the organisations that use it.

Built for Indonesia, in Bahasa Indonesia

Most of the world's prominent AI systems have been developed primarily around English-language information and globally oriented use cases. Yet the documents, administrative procedures, and everyday operations of Indonesian institutions have their own linguistic and operational requirements.

Leo-Regulus 1.0 puts Bahasa Indonesia at the centre of its development.

The model family is designed to understand and generate Indonesian text across formal correspondence, government documentation, institutional reports, financial materials, and everyday professional communication.

Kuartal is also developing support for Indonesia's regional languages and terminology, beginning with languages such as Sasak, Sumbawa, Bima, Javanese, and Sundanese. Support for regional vocabulary and less-represented dialects remains best-effort and will improve through further training and evaluation.

This linguistic foundation is intended to support every specialised AI system Kuartal develops.

Whether assisting a government employee with administrative correspondence or helping an analyst review a financial document, the objective is the same: to make artificial intelligence more useful in the language and context in which Indonesians already work.

Intelligence That Understands Your Documents

A central capability of Leo-Regulus 1.0 is its ability to work with information already held by an organisation.

Government agencies, banks, universities, and businesses maintain extensive collections of correspondence, regulations, contracts, financial statements, scanned records, and administrative documents.

Much of this information is difficult to search and often requires considerable manual effort to interpret.

Leo-Regulus combines language models with specialised document-processing and retrieval components to help organisations make better use of their institutional knowledge.

The system is designed to read digital documents and scanned materials, extract information from forms and tables, summarise lengthy reports, compare records, and answer questions using an organisation's own reference material.

Through document retrieval, responses can be accompanied by citations to relevant source passages, allowing staff to review and verify the information behind an answer.

The system can also assist with drafting official letters, preparing administrative summaries, organising records, and following step-by-step reasoning over documents and structured data.

Speech recognition and synthesis components extend these capabilities to voice-based interactions, supporting applications such as internal assistants, service counters, and institutional helpdesks.

Together, these capabilities form a practical foundation for AI-assisted work, with human review remaining essential for consequential decisions.

Introducing the Kuartal Cortex Architecture

Leo-Regulus 1.0 is not intended to function as a single, monolithic AI model.

Instead, its architecture brings together specialised models responsible for different aspects of a task.

The shared foundation covers six core capabilities: language understanding, information retrieval, speech, vision and document processing, task routing, and safety.

A routing model directs requests to the appropriate specialist, while retrieval components provide relevant institutional information and safety mechanisms help enforce operational policies.

These components form the foundation of Kuartal Cortex, Kuartal's family of industry-specialised AI systems.

Each Cortex is designed around a particular field of work. It begins with suitable open-weight models, is further trained for an industry's requirements, and can then be adapted to an individual organisation's documents, procedures, and software.

The first phase of Kuartal's Cortex roadmap focuses on four areas:

  • Government Cortex: Administrative correspondence, permit processing, institutional archives, public information, and citizen services.

  • Research Cortex: Financial research, market information, company filings, and investment reporting.

  • Due Diligence Cortex: Reviewing transaction documents, identifying potential concerns, and preparing due-diligence materials.

  • Banking Cortex: Customer documentation, credit memorandum drafting, regulatory information retrieval, and internal administrative workflows.

Additional Cortex systems are planned for healthcare, legal services, family offices, energy, cybersecurity, education, hospitality, climate, and other specialised sectors.

Rather than attempting to make every model equally capable at every task, Kuartal's approach concentrates development on clearly defined operational responsibilities.

AI That Works Inside Your Existing Software

Kuartal's approach extends beyond conversational AI.

Through Harness, the agent runtime that supports each Cortex, specialised models can be connected to the applications, databases, documents, and workflows an organisation already uses.

These connections are designed to operate through controlled integrations, including APIs, software development kits, and Model Context Protocol (MCP) servers.

This means an organisation does not necessarily need to replace its existing systems to introduce AI.

A government Cortex, for example, could be configured to retrieve information from an administrative archive, prepare correspondence, or assist with routing an application through an existing government information system.

Actions can be subject to defined permissions, safety checks, logging, and human oversight.

Kuartal Bedrock provides the operating environment and interfaces through which people interact with these systems, including administrative consoles, research dashboards, document workspaces, kiosks, and voice-based applications.

The objective is to move from AI that merely answers questions towards intelligence that can assist with actual work within an organisation's existing operational environment.

Your Data Stays Inside Your Walls

Data sovereignty is a defining principle of Leo-Regulus 1.0.

Organisations should not have to send confidential documents, internal correspondence, or sensitive operational information to an external AI provider simply to benefit from artificial intelligence.

Leo-Regulus is designed to support fully on-premises and air-gapped deployment, allowing its models and supporting components to operate on infrastructure controlled by the organisation.

In a properly configured offline deployment, document processing, information retrieval, model inference, and internal workflows can take place without an internet connection.

This makes the architecture particularly relevant for institutions with strict confidentiality, data residency, or operational continuity requirements.

Kuartal's deployment approach also supports managed environments for organisations that prefer dedicated infrastructure operated by Kuartal.

Built-in safety components are designed to help screen requests and responses, apply organisational policies, and identify sensitive information. These mechanisms operate alongside conventional infrastructure security, permissions, and human review.

Because Leo-Regulus builds upon open-weight foundation models, organisations can also retain and operate model deployments under the terms of the applicable licences.

The underlying principle remains consistent across deployment options: organisations should have meaningful control over their intelligence, data, and computing environment.

From Edge Devices to Data Centres

Leo-Regulus 1.0 is being developed with multiple deployment sizes rather than a single model configuration.

Its principal language and industry models span approximately 4 billion to 32 billion parameters, complemented by smaller specialist models for functions such as document retrieval, speech processing, and task routing.

Smaller configurations are intended for edge devices and constrained computing environments. Larger configurations are designed for dedicated workstations, GPU servers, and enterprise data-centre infrastructure.

For the Lombok Barat pilot, Kuartal has selected a 9-billion-parameter configuration running on two consumer-grade GPUs, reflecting its focus on practical and accessible local deployment.

Larger institutional deployments may require additional memory, GPU capacity, and supporting infrastructure according to their workloads, number of users, and operational requirements.

Kuartal's architecture is designed to accommodate environments ranging from individual offices and remote field sites to government and enterprise data centres.

The objective is not to require every organisation to purchase the most powerful hardware available, but to match the intelligence system to the work it needs to perform.

Developed Through Agentic Satellite

Behind Leo-Regulus is Agentic Satellite, Kuartal's internal model training and development infrastructure.

Agentic Satellite provides the framework for selecting capable open-weight models, adapting them to specific industries, and developing specialised components for different tasks.

Rather than attempting to recreate frontier foundation models from scratch, Kuartal builds upon existing model research and concentrates its development efforts on localisation, specialisation, integration, and deployment.

Each Cortex can subsequently be refined for the organisation using it, with its training process informed by relevant documents, approved data, institutional procedures, and the software it needs to operate.

Kuartal intends to continue evaluating and improving these models as new foundation models become available and additional domain-specific requirements emerge.

For Leo-Regulus 1.0, the initial emphasis remains on language understanding, reliable document workflows, information retrieval, and practical institutional applications.

The release is not positioned as a frontier coding or general-reasoning model. Kuartal will also withhold comparative benchmark claims until its own evaluations, particularly for Bahasa Indonesia and Indonesian institutional tasks, have been completed.

The First Chapter of Kuartal Intelligence

Leo-Regulus 1.0 establishes the first release in Kuartal's planned astronomical naming system for its AI generations.

Future general AI releases will follow constellation and star names, beginning with Leo-Denebola 1.5, while individual Cortex systems retain their industry-specific identities.

The naming system reflects an underlying architectural idea: many specialised forms of intelligence, each serving a defined purpose, operating together within a larger system.

Kuartal's longer-term objective is to develop sovereign AI systems for Indonesian institutions across a broad range of industries, supported by local infrastructure, open-weight models, and continuing research.

For organisations adopting artificial intelligence, the central question should not only be how capable a model is, but whether it understands their work, integrates into their operations, and remains under their control.

Leo-Regulus 1.0 represents Kuartal's first step towards answering that question.

Built for Indonesia. Trained for your industry. Operated on your terms.