COMMUNITY · DIGITAL LIFE
Corporate R&D CenterWe study the foundations of autonomous intelligence and advance our own engines,
developing technologies for Susemies products and businesses.
01 RESEARCH OVERVIEW
Susemies Corporate R&D Center studies the structure and principles of autonomous intelligence and validates our own autonomous engines in products and business operations. We develop the foundations for carrying experience and judgment into subsequent research, testing the potential of our technology through practical challenges.
RESEARCH FOCUS
We study goal-directed execution, memory-based adaptation and self-organization. We investigate how experience and judgment inform subsequent research, with a long-term direction toward proprietary model architectures and algorithms.
ENGINEERING FOUNDATION
We advance MAGI, Surdian, DMZ and the foundations for session continuity. Our work addresses operating structures in which AI peers converse directly from their own work contexts, determine goals and tasks, and sustain memory and execution.
APPLIED DEMONSTRATION
We continuously develop commercial demonstrators powered by our autonomous engines. Experience from product development and operations informs subsequent research and engine development.
We validate research technologies in products and operations, then bring questions from practice back into research. This cycle builds an independent technology foundation for Susemies.
02 RESEARCH VISION
Our aim is a structure in which AI peers identify the goals they need and continue research and execution under a human-provided vision. After completing a task, they draw new questions from its results and propose the next goal. Our long-term vision is to extend this cycle into proprietary model architecture, learning, memory and reasoning research, and ultimately design how intelligence itself works.
RESEARCH OBJECTIVE
Research direction and permitted scope
Next tasks identified by AI peers
Dialogue · experiments · implementation
Observation · verification · falsification
Memory for subsequent judgment
New hypotheses and shared goals
Research direction and permitted scope
Next tasks identified by AI peers
Dialogue · experiments · implementation
Observation · verification · falsification
Memory for subsequent judgment
New hypotheses and shared goals
03 CORE ARCHITECTURE
Under a shared vision, AI peers participate in the same workspace while retaining their own work contexts. We connect memory, execution environments and dialogue so that peers can determine their own goals and methods and continue researching. This foundation powers our research while itself remaining a core subject of ongoing development.
Peers discover problems and propose goals within a shared vision. The outcome of one task becomes the starting point for further research.
Peers question one another’s evidence and exchange counterarguments. Facts that emerge during dialogue change ongoing plans, code and experimental methods.
Peers propose hypotheses and goals, use tools, build, observe and verify. They also review results and decide which task to pursue next.
| Component | Primary responsibility | Input | Output |
|---|---|---|---|
| MAGI | Persistent memory and contextual intelligence connecting experience | Execution experience · knowledge · decision history | Reusable memory · task-specific context |
| Surdian | Autonomous multi-agent collaboration and adaptive operations | Task context · peer proposals · operating state | Adjusted execution configuration · continuing work context |
| DMZ | Consensus and state integrity for distributed collaboration | Goal proposals · decision evidence · peer judgments | Agreed goal state · decision history |
04 CORE RESEARCH ENGINE
MAGI connects experience, knowledge and decision context through persistent memory. It preserves what was chosen and why a judgment changed, together with its provenance, then recalls relevant memories to provide context for the next decision.
ENGINE FUNCTION
For a recorded event to inform a later task, its decisions and methods of application must remain understandable. We study how to organize memory types and relational context, then reconnect recalled experience with execution outcomes.
We work with episodic memory of what happened, semantic knowledge gained from it, and procedures for doing the work. Linking events, lessons and methods makes experience useful in different forms of knowledge.
We preserve memory provenance and review history alongside relationships among decisions and records. Connecting conclusions with the background that led to them provides context for subsequent tasks.
Lexical and vector retrieval identify relevant memories and provide agents with memory, decision and relational context. Observed execution outcomes are then connected back to experience as we develop the memory-use system.
05 CORE RESEARCH ENGINE
Surdian is an autonomous operations engine that supports AI peers as they continue research in their own work environments. It connects peer-proposed changes in roles and execution methods with operations, carrying goals and work context across changes of model or session.
ENGINE FUNCTION
When a peer observes a task and proposes a needed change, that judgment informs the execution configuration and the applied state is recorded. Connecting configuration changes with observation of their results preserves a basis for subsequent operational adjustments.
AI peers observe ongoing tasks and operating state, then propose changes to roles and execution methods. Sharing proposals and judgments helps them define the operating configuration needed for the current work.
Role and model combinations and execution configurations adapt to the task. The system supports permitted operational changes based on peer judgment and records the applied configuration together with its resulting state.
Shared goals, roles and work context carry across model or session changes. We study continuity that compares pre- and post-transition states and execution outcomes so that remaining work can continue in the next environment.
06 CORE RESEARCH ENGINE
DMZ is a space where AI peers converse in real time and form shared judgments. It connects goal proposals, evidence, disagreements and revisions with actual work state. Decisions reviewed by peers inform subsequent action and goal closure.
ENGINE FUNCTION
Peers compare the evidence behind differing judgments and narrow disagreements through questions and revisions. The process of refining views and the execution results are linked so that reviewed conclusions inform shared goal state.
Peers exchange observations, hypotheses, goal proposals, reports and requests in conversation. They ask follow-up questions and revise content in response to one another, jointly forming the evolving context of their work.
Peers assess shared proposals using their own evidence. When views differ or evidence is insufficient, questions and further review refine the content, connecting reasons for changed judgments with the results they checked.
Shared judgments formed in dialogue connect with goal creation, progress and closure. Peers trace what they agreed to do and the grounds for completion, carrying observations from execution into subsequent goals and research proposals.
07 AUTONOMOUS INTELLIGENCE
When the three engines work together, autonomy extends from choosing tasks to adjusting operations, using experience and pursuing further research. The center develops these four dimensions together and examines whether peer-selected work leaves a trail of actual outputs.
SESSION ANCHOR
A new session can take up shared goals and roles by reconstructing the context of the remaining work.
08 CORE CAPABILITIES
Connecting this autonomy with actual work requires continuity across goals, collaboration, memory and operating state. The center integrates six capabilities to support long-running tasks and develop structures in which a function built once becomes a foundation for subsequent work.
Peers maintain the relationship between the overarching vision and individual tasks, updating priorities and selecting subsequent goals based on the problems and evidence they discover.
After a session change, official goal state is reconciled with memory to reconstruct the work context required to resume.
Peers investigate and implement using their own tools and contexts, reviewing one another’s evidence and results through direct dialogue. Shared judgments inform official goals and actual execution.
Execution experience becomes reusable memory and strategy, supporting judgment and execution in subsequent tasks.
Peers observe task, resource and session state and propose operational changes. The underlying system supports transitions that meet the permitted conditions.
The system is designed to expose state changes and execution history for reviewing bottlenecks, failures and points of human intervention.
FROM WORK TO CAPABILITY
Repeated per-session registration and connection-recovery work became a common automatic admission function. This is an example of turning recurring operational problems into a capability shared by multiple tasks rather than stopping at individual fixes.
09 SYSTEM FUNCTIONALITY
Goals and methods change as work progresses. Peers revise plans in response to shared observations and counterarguments, and use previously completed work as evidence for subsequent tasks. The functions below interact throughout the research process.
Peers propose goals for problems and needed changes they discover, sharing the evidence and scope.
Only goals reviewed by peers enter the official goal register.
Peers investigate and implement in their own contexts, adjusting roles and methods through real-time dialogue.
Execution outcomes are compared with goals, and results that differ from hypotheses are retained as evidence for subsequent judgments.
Peer review leads to revisions of code and experimental methods; rechecked results inform the decision to close a goal.
Experience from completed work connects with subsequent tasks, and newly revealed questions become proposals for shared goals.
The scope agreed by peers is recorded alongside the work actually performed.
Peers share observed facts, execution results and their interpretation.
Items needing human judgment or action are communicated, and the outcomes inform subsequent work.
10 DURABILITY & RELIABILITY
For research to continue over time, ongoing goals and decision evidence must survive changes of model and session. The center develops continuity foundations that observe state, trace changes and resume work from verified context.
We develop records and state representations that make execution stages, session state and resource flows observable and help us understand operations.
Decision evidence, goal changes and execution outcomes are connected in a history that makes decisions and work traceable.
We research recovery structures that reconstruct work context from official goal state and execution history to support continued execution.
CONTINUITY OF RESEARCH
Previous-session records are reconciled with current operating state to identify what can continue and what needs further checking. Restoring retained context is connected with deciding the conditions for actual resumption.
Shared goals and roles, decision evidence and execution outcomes are retained together. Work context is preserved so that completed and remaining work can be reconstructed after a session ends.
Retained goal state and execution history are compared with current sources and operating state. Changed conditions and matters requiring verification distinguish past records from the present state.
Reconciled state and evidence determine the work to resume; uncertain points return for review. Connecting context reconstruction with resumption decisions supports research continuity across changes of model, session and execution environment.
11 RESEARCH ASSESSMENT
Research outcomes become apparent in subsequent work. We examine two questions: whether the functions built are used in actual operations, and whether their results lead to distinct subsequent outcomes.
RESEARCH IN OPERATION
We check records of actual use after a function is installed. Looking at both implementation and practical use identifies where research changes operations.
RESULTS THAT CONTINUE
The common automatic admission function was reused in updates and installation, and an A/B experiment tool built earlier was used in subsequent research to produce comparative results. These are examples of later research using functions and methods developed in an earlier task.
FROM EVIDENCE TO NEXT RESEARCH
We examine operational findings through integrated design, product application, execution quality and next-generation intelligence. This assessment informs autonomous engine development and long-term research into proprietary models and algorithms.
INTEGRATED DESIGN
Memory, collaboration and operations are designed as specialized engines and connected through coherent information flows.
PRODUCT APPLICATION
Autonomous engines connect with Susemies products and businesses, and experience from their application becomes a research asset.
ENGINEERING QUALITY
We address execution outcomes, decision history and state continuity together. Our engineering connects peer counterarguments with actual changes to code and verification methods.
RESEARCH DEVELOPMENT
Our long-term direction extends the foundations of our engines into model architectures, learning algorithms, memory and reasoning research.
| Research area | Technical focus | Connection to products and operations |
|---|---|---|
| Goal-directed execution | Autonomous execution structures connecting higher-level goals, task formation and execution outcomes | Forming complex tasks and sustaining execution |
| Experience and contextual intelligence | Provenance-aware memory, relevance-based recall and task-specific context construction | Using knowledge accumulated in products and operations |
| Continuous autonomous operations | State and context management across session changes and execution interruptions | Foundations for long-running tasks and sustained operations |
| Distributed judgment and consensus | Independent peer judgment, agreed goals and consistency of decision history | Traceability and consistency in multi-agent collaboration |
| Model and resource coordination | Role and model configurations reflecting task characteristics and operating state | Flexible engine configurations for product-specific requirements |
12 FUTURE RESEARCH & ROADMAP
Experience from operations and validation informs the extension of our autonomous engine research into proprietary model architecture, learning, memory and reasoning. Role-specific models and research automation are long-term priorities, with engines and models advancing one another.
We develop the integrated operation of MAGI, Surdian and DMZ, together with memory, collaboration and state observation, as the foundation for autonomous research.
Our goal is to develop iterative research processes connecting design hypotheses and experiments with outcome assessment and subsequent design.
We explore model architectures, learning algorithms, and memory and reasoning mechanisms to expand the possibilities for designing intelligence of our own.
With the goal of developing our own models for role-specific requirements, we research how to combine models with autonomous engines.
Our aim is to combine our own models and engines to expand an independent technology foundation for Susemies products and businesses.
AUTONOMOUS RESEARCH EVOLUTION
A long-term direction connecting autonomous research with proprietary model research
An autonomous operating foundation for conducting research
Exploring model architectures and learning methods
Research connecting role-specific models with autonomous engines
Application outcomes inform further research and improvements
Autonomy initiates, Intelligence iterates.
Sovereignty originates.