Join Us
Projects

Real-world testbeds for agent-native systems.

Experimental systems for validating agent-native execution, orchestration, trust, and real-world impact.

01Active

Agentivium Core

Shared schemas, SDKs, protocols, and design principles for building agent-native systems across domains.

Agentivium Core defines the shared foundations of our agent-native stack: Digital Profile, Digital Stream, Evidence Packet, Adaptive Action, and Evaluation Protocol.

Research Themes

Digital ProfileDigital StreamEvidence PacketAdaptive ActionEvaluation ProtocolShared schemas and SDKs

Expected Artifacts

  • Core schemas
  • Python and TypeScript packages
  • Validation CLI
  • Evaluation protocol
  • Digital Entity Loop diagram
02Prototype

Learner Intelligence Layer

A learning testbed where learner profiles, learning streams, and evidence-based recommendations validate the Digital Entity Loop.

This testbed applies Agentivium Core to education by modeling learners as evolving digital profiles updated from learning events and used for evidence-based exercise recommendations.

Research Themes

Learner Digital ProfileLearning StreamEvidence-based recommendationTeacher usefulnessProfile freshness and completeness

Expected Artifacts

  • Learner profile schema
  • Learning event taxonomy
  • Exercise recommendation baseline
  • Teacher usefulness rubric
03Prototype

Xanh Narrative Engine

An agriculture testbed that converts plant, product, and farm traces into evidence-grounded narratives.

Xanh Narrative Engine validates the Digital Entity Loop in agriculture by connecting plant/product profiles, farm event streams, evidence packets, and narrative generation.

Research Themes

Plant/Product/Farm ProfileFarm/Product StreamEvidence-grounded storytellingTrace-to-narrativeDomain-grounded evaluation

Expected Artifacts

  • Plant/product event taxonomy
  • Trace-to-narrative pattern
  • Evidence-grounded narrative schema
  • Narrative evaluation rubric
04Exploratory

hpc-claw

An infrastructure testbed for compute-aware orchestration of agentic workloads on HPC systems.

hpc-claw treats agentic workloads as schedulable, observable, and governable workloads, contributing execution streams, agent action traces, and value-per-compute metrics back to Agentivium Core.

Research Themes

Execution StreamAgent Action TraceCompute-aware orchestrationBudgeted executionValue-per-compute

Expected Artifacts

  • Execution event schema
  • Agent workload trace schema
  • Budgeted orchestration baseline
  • Value-per-compute metrics
05Exploratory

Smart Building PoC

A cyber-physical validation of policy-aware agent-native execution using room, device, and sensor streams.

The Smart Building PoC applies Agentivium Core to room and device profiles, sensor/device streams, evidence packets, and policy-aware suggested actions.

Research Themes

Room/Device ProfileSensor/Device StreamPolicy EnvelopeHuman approval loopCyber-physical safety

Expected Artifacts

  • Cyber-physical event schema
  • Device/action policy schema
  • Human approval workflow
  • Safety edge cases
Contribute

Interested in building agent-native testbeds?

Join UsResearch Areas