ADVANCED AI RESEARCH AND ENGINEERING · TOKYO
Sovereign AI. Persistent agents. One runtime.
KodaSōken builds specialist SLMs and governed multi-model systems for high-trust industries. Local control. Frontier capability. One identity across every model.
EXPLORE: Sovereign Models · KoLo Agents · KoLo Runtime · Research
One identity. Many minds.
01 — WHAT WE BUILD · THE KODA STACK
The complete operating stack for sovereign AI.
Most AI products begin and end with a model. KodaSōken develops the models, agents, runtime and research infrastructure required to operate specialist intelligence continuously and under governance.
KODA-owned SLMs and mid-scale models for private, local, edge and domain-specific work.
Long-running operating identities that preserve authorised memory, role, policy and unresolved work across sessions and model changes.
The operating layer that controls identity, memory, model routing, evidence, tools, permissions, audit and recovery.
Project Nyx, Identity Fidelity, multi-model cognition, edge intelligence and continuity across future models and computing substrates.
02 — HOW IT WORKS · MULTI-MODEL INTELLIGENCE
One agent. The right mind for every task.
A super-agent is a persistent agent that keeps identity, memory and policy across sessions and model changes. It is not permanently tied to one model — it preserves its operating state outside the weights, then recruits the appropriate intelligence for each task. One identity. Many minds.
KoLo coordinates models, evidence, tools and human authority within one accountable operating workflow.
1 — Handle routine work locally
Compact SLMs perform classification, extraction, routing and frequent specialised tasks close to the institution.
2 — Recruit specialist intelligence
KODA specialist models perform professional and sector-specific work using authorised evidence.
3 — Escalate only when necessary
Approved frontier models contribute when a task requires broader knowledge or deeper reasoning.
4 — Verify before consequential action
Guardian systems, deterministic checks and authorised humans review the result according to risk and policy.
KoLo records which model contributed, which evidence was used, which tools were accessed, which policy applied and who approved the outcome. The agent is the continuity. The models are the minds it recruits.
See the complete architecture →
03 — SOVEREIGN AND FRONTIER INTELLIGENCE · CONTROL WITHOUT ISOLATION
Own what must stay. Recruit what must scale.
KODA develops sovereign models for capabilities that require local operation, privacy, low latency, predictable cost, institutional control, offline resilience or sector-specific language. KoLo also recruits approved frontier models when broader capability is genuinely needed.
Sovereign where control matters
Models, adapters, data manifests, evaluation and deployment remain under defined ownership.
Frontier where capability helps
External intelligence is recruited for approved tasks using the minimum necessary context.
Governed throughout
Identity, evidence, permissions and accountability remain inside KoLo.
Organisations do not have to choose between complete external dependence and complete technological isolation.
Explore the KODA model family →04 — INDUSTRIES · SPECIALISED INTELLIGENCE
Built for work where context, privacy and accountability matter.
Clinical documentation, handover, patient communication and healthcare operations under human clinical authority.
Manufacturing & industrial AI →
Maintenance, quality, technical documentation, inspection and safety through edge-deployable intelligence.
Education & professional learning →
Persistent tutoring, curriculum-grounded assistance, simulation and teacher-controlled learning systems.
Skills, language preparation, documentation, onboarding and continuing worker support across countries.
These sectors are examples, not boundaries. KODA can adapt the same architecture to other industries requiring specialist models, local control, persistent context or governed action.
Explore all industries →
05 — APPLIED SYSTEMS · THE ARCHITECTURE IN REAL WORK
Research becomes valuable when it enters an operating environment.
KODA Kenkō · Active development
Sovereign healthcare intelligence combining medical models, persistent agents, controlled clinical evidence and professional approval.
KODA FLOW · Controlled-pilot pathway
Hospital operations intelligence for patient flow, bed coordination, transfers, discharge and institutional visibility.
KIP Dev · Operational development system
A persistent multi-model development environment for planning, implementation, validation and project continuity.
Yoko & KipTalk · Active product development
Mobility guidance, professional-language learning, cultural preparation and long-term worker support.
06 — RESEARCH AND EVIDENCE · MEASURED CONTINUITY
Advanced AI research tested in running systems.
KodaSōken researches how artificial systems preserve identity, memory, role and accountability while their models and infrastructure change — Project Nyx, the Identity Fidelity benchmark, KoLo telemetry and the publication programme.
F = 0.961
Latest Identity Fidelity result — live substrate swap.
4 transitions
Verified model transitions with continuity preserved.
336+ versions
Current versioned-memory lineage, none lost.
Internal measurement unless otherwise marked. External review and reproduction status are shown with every result.
間
Build the intelligence you need. Keep control of how it operates.
KodaSōken works with research institutions, hospitals, industrial operators, education organisations, model developers and compute partners. Start with one operating problem, one defined boundary and one measurable result. Sovereign models. Persistent agents. Governed by KoLo.
“Dream Less, Act More.”
Partner with KODA →