HPC & BioCompute Lab
Continuity middleware for future biological compute.
KoLo is a substrate-portable runtime designed to preserve memory, task state, governance, and observability as AI agents move across today’s model substrates and future biological or neuromorphic compute interfaces. We measure what persistence costs, and what it buys.
The thesis
Cognitive substrates will change. Runtime continuity must be measured.
Biological-compute platforms, organoid arrays, and neuromorphic chips are moving from research into programmable infrastructure. What remains open is the software layer around them: memory persistence, task-state recovery, auditability, closed-loop scheduling, and substrate abstraction.
KodaSōken is preparing that layer. The HPC & BioCompute Lab is where we design the benchmarks, simulation protocols, and integration methods needed to test this architecture with future partners.
Research tracks
Four tracks. One continuity architecture.
Continuity benchmarks
Measurable persistence, recovery, memory, and role continuity across substrate changes. 68:1 autonomic-to- cognitive runtime ratio as the first KoLo reference benchmark.
HPC simulation
Large-scale agent mesh, substrate-switching tests, and stress experiments at lab scale. Reproducible runs across LLM providers, edge devices, and simulated biological substrates.
Biological-compute middleware
Memory, orchestration, logs, and APIs designed to wrap future neural substrates. The boring infrastructure that lets biological compute become programmable, observable, and recoverable.
Future partner experiments
Protocols for universities, biological-compute labs, chip/HPC partners, and medical research institutes. The objective is to support reproducible benchmarks, shared telemetry, and publishable research once formal collaborations are established.
Tested today
Already running. Already measurable.
These properties are observable on the running KoLo runtime against current LLM substrates. Telemetry samples shared under partner NDA.
LLM substrate switching
Live transitions across Anthropic, DeepSeek, Qwen, and OpenAI Codex with continuity state preserved.
Agent mesh simulation
Five-process peer mesh running under active research conditions. A2A protocol with type-safe ordered messaging across substrates.
Memory-state recovery
Versioned persistent memory. 139+ snapshots accumulated without loss across sessions, models, and providers.
Daemon / autonomic runtime cycles
144 closed-loop cycles per process per day at zero token cost. 13/13 daemon module tests passing since March 2026.
Failure and outage recovery
Predefined continuity-preservation routines validated under degraded network and provider-instability conditions (April 17, 2026 security-stress event).
Telemetry and audit logs
Append-only event ledger. Memory writes, daemon events, and cross-process messages are recorded in replayable, inspectable, exportable logs.
Requires partners
Still ahead. Designed to test with partners.
These items cannot be validated by KodaSōken alone. They are on the lab’s road map for substrate-partner engagement.
Biological neural substrate access
Lab access to wetware compute or cultured neural arrays with read/write interfaces suitable for closed-loop experiments.
Wetware / biocompute API integration
Substrate-adapter implementation against partner hardware, including handshake, scheduling, error model, and observability bridges.
Closed-loop stimulation / response testing
Reference experiments demonstrating the daemon’s closed-loop scheduling driving partner-substrate stimulation and reading partner-substrate response under governed protocols.
Energy and latency profiling
Substrate-specific energy-per-operation and end-to-end latency measurements, recorded against the continuity benchmark for partner-specific deployment profiles.
Biological continuity benchmark design
Partner-ready methodology for measuring memory, role, and state-continuity preservation when execution moves from silicon toward biological neural compute. Publication potential subject to formal collaboration.
This page describes our research direction and the partner profile we are preparing for; formal collaborations will be announced only when established.
Who this is for
Three partner profiles.
- Biological-compute platformsWetware, organoid, or neural-substrate compute providers exploring runtime, continuity, and benchmark layers around their hardware.
- HPC & chip partnersNeuromorphic, edge-accelerated, and HPC providers interested in large-scale agent-mesh simulation, substrate-switching stress tests, and reference topologies for distributed neural compute.
- Research institutionsUniversities and medical research centres interested in continuity benchmarks, biological-compute readiness, and governance frameworks.
Explore BioCompute integration research.
We are preparing KoLo for future biological-compute, neuromorphic, and HPC substrate testing.
We welcome discussions with research and infrastructure partners.