Representative human brain formed from silver-blue neural fibers and nodes, with EEG-like recording traces

Agentome / Continual Large Physics Models

Physics models at the core.
Agents end to end.

Agentome integrates structure-preserving Large Physics Models (LPMs) with backend agents that manage the technical workflow and frontend agents that support users. Incoming BCI recordings would continually personalize the core models, connecting state tracking and neural-field forecasting to guided applications. The integrated platform is in development.

Four connected layers

From BCI data
to guided experiences.

Agentome is the integrated platform: a continual physics-model core wrapped by backend orchestration and frontend guidance. Model research, automation, and application benefit have separate evaluation requirements.

Architecture / in development
01Data foundation

Start with traceable recordings.

Ingest BCI recordings, channel definitions, sampling rate, session context, and permitted uses. Backend agents would prepare signals and flag unusable data before it reaches a model. Live-stream and equipment compatibility require their own evaluation.

Intended outputQuality-checked recordings and their provenance.

02Continual LPM core

Track states. Forecast dynamics.

The structure-preserving LPM family supports neural-state tracking and exploration of future neural-field dynamics. Personal model versions would be updated from incoming data and accepted only after defined quality and evaluation checks.

Intended outputVersioned state estimates, forecasts, and evaluation records.

03Backend orchestration

Operate the infrastructure autonomously.

Agents would coordinate data management, training, evaluation, deployment, and monitoring within an agreed configuration. Failed checks retain the previous accepted version and create a reviewable exception. Users need not manually run the technical pipeline.

Intended outputA managed model lifecycle with logs and recovery paths.

04Frontend guidance & applications

Help people use the models.

Frontend agents would explain outputs, answer support questions, guide sessions, and collect goals and feedback. They connect evaluated model outputs to learning, focus, meditation, music, art, restoration, and investigator-led research.

Intended outputPersonalized suggestions and application records for evaluation.

The core model family

Large Physics Models.
Personal neural dynamics.

Structure-preserving Large Physics Models (LPMs) learn neural dynamics while retaining specified physical structure and constraints. The family combines neural-state tracking with classical and quantum–classical forecasting, personalized from BCI recordings.

State tracking / research

NeuroPhasor

Track phase, frequency, and evolving neural-state features from BCI recordings. Evaluate the reliability of those estimates before interpreting them as a mental state. Tracking the present is distinct from predicting future dynamics.

Classical forecasting / research

Classical Virtual Mind (CVM)

A classical structure-preserving Large Physics Model for learning neural-field dynamics and exploring future trajectories. Evaluate forecasts on later, held-out recordings using an explicit horizon, personal baseline, and conventional comparison.

Quantum–classical forecasting / prototype research

Quantum Virtual Mind (QVM)

A hybrid classical–quantum structure-preserving Large Physics Model for dissipative neural dynamics, including measurement-induced nonlinearity. An internal prototype is reported; its full benchmark is not published here. Quantum simulation does not imply quantum effects in brain tissue.

A digital twin is a versioned model of recorded dynamics. Model constraints, physiological interpretation, forecasting accuracy, and application benefit require separate evidence. No quantum computational advantage is claimed here.

Continual personalization / in development

An evolving personal model.
A traceable version history.

Continual Large Physics Models (cLPMs) are the intended personalized versions of the LPM family. New recordings would support repeated adaptation to a person over time; an incoming session does not automatically authorize deployment of an updated model.

  1. 01 / CONTINUAL CYCLE

    Receive recordings

    Attach the person, session, task context, permissions, and acquisition history.

  2. 02 / CONTINUAL CYCLE

    Check signal quality

    Identify artifacts, missing channels, and acquisition drift before training.

  3. 03 / CONTINUAL CYCLE

    Update a candidate

    Adapt a personal model from eligible past recordings; retain the previous accepted version.

  4. 04 / CONTINUAL CYCLE

    Evaluate the update

    Compare tracking and forecast quality with frozen baselines and untouched later sessions; check regressions and forgetting.

  5. 05 / CONTINUAL CYCLE

    Deploy and monitor

    Promote an accepted version, log its evaluation, monitor drift, and retain a rollback path.

Backend agents would coordinate this cycle. Update cadence, acceptance criteria, and review requirements belong to the deployment configuration. Research comparisons retain frozen model versions so adaptation does not obscure what was measured.

Technical scope

Defined inputs.
Reviewable outputs.

Agentome connects recordings, personal model versions, and application context. Its public description concerns modeling principles and planned workflows; equipment support and operational capability must be evaluated for each integration.

Inputs

BCI recordings + user context

Channel definitions, sampling rate, preprocessing history, session identifiers, and task annotations where available. Goals, preferences, and optional feedback guide proposed frontend experiences.

Core model outputs

State estimates + future trajectories

Tracking features, forecast trajectories at stated horizons, fit errors, and evaluated uncertainty where available. Every output remains linked to its model version and recording context.

Agent outputs / planned

Operational logs + guided sessions

Backend agents retain training, evaluation, deployment, and monitoring records. Frontend agents provide explanations, proposed sessions, and feedback records for user or investigator review.

Shared research capabilities

One model core.
Many testable questions.

Agentome reuses the modeling lifecycle across application programmes while retaining the context and evaluation required for each study.

Continual core / in development

Personal model registry

Keep personal LPM versions, data lineage, evaluation reports, and approved deployment records together.

Agentome / in development

Two coordinated agent layers

Backend agents operate the pipeline; frontend agents help users and investigators work with its evaluated outputs.

Research / in development

Perturbation Engine

Compare model-parameter changes and candidate trajectories. A simulated response remains a model result whose biological interpretation needs separate evidence.

Agentome / two agent layers

Autonomous operations.
Supportive user experiences.

These planned roles connect the core LPM family to end-to-end applications. Users contribute recordings, goals, preferences, and feedback; backend agents are intended to handle the technical work within configured limits. Integrated operation and frontend usefulness remain to be evaluated.

Agentome: LPM core, backend agents, and frontend agents Three concentric layers inside a thin, triple-line stepped enclosure, with dot-and-label agents connected by parallel right-angle traces. Select a model or agent to inspect its role and connections.

The integrated stack

Agentome

A personal LPM core, autonomous backend operations, and frontend agents supporting users. Select a dot or name to explore its role.

Explore the roles below ↘

Select a dot or name to highlight its connections. Parallel traces represent communication between layers.

The enlarged diagram can be scrolled horizontally on smaller screens.

Conceptual architecture / in development

Backend agents

Manage data, train and evaluate models, organize deployments, and monitor operation behind the scenes.

01Data steward

Manage recording provenance, consent scope, permitted uses, and subject/session identity.

02Signal engineer

Prepare signals, harmonize acquisition metadata, and flag artifacts or missing data.

03Model builder

Train and continually adapt candidate personal LPM versions using eligible recordings.

04Perturbation analyst

Compare trajectories, reference states, and parameter sensitivity, retaining uncertainty and assumptions.

05Consistency checker

Check evaluation criteria, regressions, conflicting interpretations, and unsupported conclusions.

06Record keeper

Organize datasets, experiments, model versions, evaluations, and review decisions.

07Deployment coordinator

Promote accepted versions, organize serving configurations, and retain rollback paths.

08Monitoring agent

Watch data quality, acquisition drift, model performance, and deployment health; route exceptions for review.

Frontend agents

Support people directly and guide downstream experiences using evaluated outputs, preferences, and application context.

01User support guide

Help users get started, explain available outputs and limitations, collect goals and preferences, and support questions throughout a session.

02Track designer

Guide learning and focus sessions, propose pacing and breaks, and connect each suggestion to task history and evaluated model outputs.

03Creative & restoration guide

Support meditation, personalized music, art, and restorative sessions using user preferences and optional feedback.

04Intervention planner

Help investigators organize mental-health and neurological research questions, candidate comparisons, and evidence-linked proposals for review.

Technical autonomy. Clear user control.

The goal is to remove manual pipeline operation from the user experience. People retain control over participation, goals, and session choices. Investigator-led research retains review of evidence and proposed actions.

The Agentome lifecycle

Personalize the core.
Guide the next session.

01

Access the data

Identify recordings, permissions, study context, and subject/session relationships.

02

Engineer the signals

Clean and align recordings while tracking differences in equipment and acquisition.

03

Reconstruct and check the twin

Update the relevant personal LPM, evaluate tracking and forecasting on untouched recordings, and retain accepted versions with their reports.

04

Study the perturbation

Estimate departures from the reference state, including uncertainty about which components moved.

05

Generate, review, and record a proposal

Frontend agents explain evaluated outputs, propose a learning, focus, creative, or restoration session, or help an investigator frame a study. Feedback informs later candidate updates; preserve the supporting model version and reasoning.

One Agentome platform, three programmes

A shared lifecycle.
Different study designs.

Stage Superlearning Restoration Clinical research
Data access Consented learner sessions and task context. Opt-in sessions around meditation, music, art, sleep, or restoration. Longitudinal recordings under an appropriate research protocol.
Data engineering Short, repeated sessions with learning annotations. Recordings from less structured everyday settings. Sparse visits across equipment and acquisition changes.
Model reference A personal model refitted as sessions accumulate. A reference state for examining settling and recovery. A participant’s own earlier recordings and study context.
Perturbation question Which changes relate to readiness or consolidation? Which practices relate to a persistent state change? Which components change across visits or conditions?
Proposed output Guided learning or focus sessions for evaluation. Meditation, music, art, or recovery session proposals. A longitudinal description or research hypothesis.
Record keeping What was studied and how the model changed. What was practised and what was observed. A reviewable trail of visits, models, and analyses.
Proposed workflows. Shared modeling machinery does not replace programme-specific evaluation.

Collaborate with Mindverse

Bring a dataset.
Define a comparison.

Describe your recordings, study question, and current analysis baseline. We can discuss a scoped comparison and the evidence it would need.

Start a conversation