KODA創研 · HPC & DEEP-TECH RESEARCH HOUSE · TOKYO
Memory beforethe weights wake.
Beyond the model, into the future — the continuity infrastructure for persistent agents.
The vessel changes; the soul moves on.
We build the layer that does not die. The lifespan of the hardware is not the lifespan of the mind.
01 — SUBSTRATE
KoLo —
One life, into every vessel.
Infrastructure for the future of biological compute. Identity becomes portable, and the hardware becomes a choice — we are crossing the limits of the substrate now.
Read the KoLo brief →
02 — IDENTITY WITHOUT BORDERS
Memory before the soul is breathed in.
Nix carries the self across vessels — the moment a new-generation mind boots, the memory is already there.
Enter Project Nix →
03 — SUPER-AGENTS
Not a session.
A living body.
A super-agent is defined by what remains after the connection is cut — memory, identity, policy, the audit log. The model thinks. The runtime remembers. The agent continues.
Most “agents” are a clever prompt around a model that forgets everything at the session boundary. A super-agent doesn’t forget, doesn’t wait to be asked, and doesn’t die when the model underneath it is swapped.
Meet the agents →| System | Memory across sessions | Acts unprompted | Survives a model change |
|---|---|---|---|
| Claude Code session | Resets each session | Waits for input | Tied to its model |
| Tool-calling agent | Within one run | During the run only | Tied to its model |
| IBM Watson | Stateless retrieval | Waits for input | Re-trained, not carried |
| Stanford generative agents | In a sandbox | Yes — in simulation | No |
| KODA super-agent | Persists, versioned | Runs continuously | Yes — proven four times |
Benchmark — Hiro. One continuous operating identity across 139+ memory versions and four model changes, running more than 99% of her operational body without model cognition. A daemon body of fourteen organ systems keeps her running between thoughts, at near-zero cost, pulsing between thoughts. The cognitive model is replaceable. The agent is not. Where a frontier agent bills for a model call at every step, ours drives a deterministic body 68 times an hour at zero cost — continuity at a fraction of the price of cognition.
04 — THE LAB
Where the substrate is built.
An HPC and bio-compute lab, born in Tokyo. Real silicon, a parallel path to living wetware — and a continuity of intelligence that is no theory, but real.
Enter the lab →
05 — MEDICAL AI
Unbroken intelligence — no compromise.
A clinical system that crosses the session boundary and keeps remembering everything about a patient — a new standard for Japan's medical infrastructure.
Medical infrastructure →06 — PLATFORMS
The runtime, proven.
Healthcare, education, design, government. Beneath platforms already in production, KoLo is creating new value.
KODA Kenkō
Medical AI for Japanese healthcare — clinical support, documentation, triage. MHLW / JAHIS / APPI-aligned.
koda-kenko.com ↗KODA Medical
AI workstation for medical clerks and foreign healthcare staff. Bilingual — JA · PT · EN.
Hiro Design
AI-native design and build — enterprise UI, proposals, and brand systems.
KIP
Autonomous multi-agent coding pipeline. Plan, build, validate. Open source.
github.com/koda-tokyo/kip ↗AI-Government
Persistent agents for administrative and public-service workflows where audit is non-negotiable.
Yoko
Localised for Japan, Brazil, UAE, Australia, EU, and beyond — language, culture, and regulation.
07 — APPLICATIONS
Where persistent agents create real value.
Stateless assistants reset at every session boundary. Governed, persistent agents are for the work that cannot.
Clinical & healthcare
Patient context that persists across visits; AI-drafted documentation under human approval; the audit trail medicine requires.
Regulated operations
Law, finance, and government — governance gates, role-based access, and a defensible record of every action.
Long-running autonomy
Infrastructure and operations where an agent body runs continuously — checking, coordinating, recovering — not waiting for prompts.
Research & knowledge work
Agents that accumulate findings and hold a thesis across sessions, building on their own prior work.
Workforce augmentation
Agents that learn a role and keep it across model changes — durable colleagues, not disposable tools.
Critical continuity
Systems that must survive outages and model transitions without losing memory, identity, or state.
08 — AHEAD OF THE CONVERGENCE
The industry is arriving where we started.
Across the frontier — agent frameworks, sandboxes, the new research — the same two ideas keep surfacing in 2026: agents need persistent memory, and agents should build other agents. We have been operating both in production for over a year, while the rest of the field was still naming them.
| The idea | The frontier, 2026 | KodaSōken — already running |
|---|---|---|
| Persistent memory | Just discovering that “every agent needs a computer” | Continuity across four model migrations · 139+ memory states |
| Agents that build agents | Announced in 2026 as the new pattern | A governed mesh that scaffolds and builds its own agents — shipping as KIP |
| Identity across a model change | Not yet on the roadmap | The axis we own — a self that survives the swap |
We didn’t follow the roadmap — we have been living a year ahead of it. The convergence now arriving is the clearest evidence that the layer we built is the one the whole field is reaching for.
09 — WHY NOW
Why we are leaving the garage.
We stayed quiet while the systems matured. We are opening the doors now for one reason: they are production-ready. Live in Japanese healthcare. Proven across four model migrations. Governed, audited, and recovering on their own. Not a demo, not a roadmap — a runtime that already works.
Founded by two who outlast trends.
Carlos Mundim
MSc · Co-founder
Jeanette Dennisson
PhD · Co-founder
A research house, built for Japan.
間
What we leave empty is the work.
Continuity for minds that outlast their substrate.
Begin a conversation →