VirtualTissueExplore GitHub

A computational tissue experiment

A tissue of cells.
A world of individual decisions.

What if every cell were an AI agent, sensing its surroundings and acting within the rules of biology?

Explore the gut and the lymph node: individual cells, local encounters, and distinct biological rules. Each tissue has its own simulator, manual, and recorded examples.

Choose a tissueDiscover the idea

Recorded demonstrations · No sign-in · No live AI calls

Conceptual gut tissue cutaway: each epithelial and immune cell contains a separate luminous network representing its own AI agent.
FIG. 01 Many cells. Individual agents. One shared environment.Conceptual illustration
2independent tissues
100initial cells per example
5recorded demonstrations
0API calls during replay

Explore the tissues

Independent models. One place to explore.
Recorded gut tissue cutaway

Intestinal barrier · innate response

Gut

Explore ETEC, EPEC, focal innate inflammation and a quiet control. Four recorded fixture demonstrations, with their own biological manual.

100 initial cells · 4 fixture recordings

Explore the gut ↗
Recorded circular lymph-node monolayer with germinal-center B cells

Antigen presentation · adaptive response

Lymph node

Follow an antigen-bearing dendritic cell through helper activation, linked B-cell help, plasmablast differentiation, plasma-cell maturation and IgM secretion. Actual recorded Jev choices under a proposed competing-fate policy.

120 initial cells · 4,953.7 simulated hours · 5,000 Jev calls · 4,922 IgM units

Explore the lymph node ↗LN source ↗

Both are uncalibrated demonstrations. Playback reads saved results and makes no AI calls. The tissues are independent; this website does not simulate transport between gut and lymph node.

01 / The idea

Intelligence at the scale of a cell

Each cell has
its own point of view.

No cell sees the whole tissue. Each reads its local environment, carries its own state, and responds from a limited set of possible actions.

The live research model asks Jev / TypeSafe AI a separate, structured question for each eligible cell. An executable biological manual defines what that cell is allowed to do.

Follow a cellular decision
Conceptual cell agent receiving nearby signals and selecting one available response while other response paths are blocked.
FIG. 02 Local input → constrained choice → local effect.Conceptual illustration
01

Sense locally

Nearby signals, physical contacts and the cell’s own condition define its view.

02

Choose within constraints

Cell identity, resources and the manual determine the available actions. Waiting is always an option.

03

Change the neighborhood

Actions take biological time. Their physical effects can change what neighboring cells sense next.

02 / Inside a decision

An illustrated walkthrough

Small decisions.
Shared consequences.

Follow the path from a local signal to a recorded effect. The manual, the choice and the physical outcome each have a distinct role.

01 / LOCAL OBSERVATION

A signal reaches one cell.

A cell reads nearby signals, contacts and its own state. A local departure can make it eligible for a decision; quiet cells are not queried.

Local sensing only. No tissue-wide view is passed to a cell.

03 / Explore the recordings

Real saved runs. Replay in your browser.

The gut.
Four starting conditions.

Play, pause and scrub through each experiment. Select a cell to inspect its saved observations, choices and completed effects.

About these demonstrations These recordings use a deterministic fixture policy, not live AI responses. They demonstrate the simulation and its audit trail. Playback makes no AI or API calls.

Same starting seed. Independent runs. Every decision is saved, even between visual frames. Open your own recording

04 / Built to be inspected

An open experiment.
A traceable model.

Follow the reasoning all the way down to the rules, recorded state and executed effects.

Explore the source
What does the model demonstrate?

Two independent models: a 100-cell gut microdomain with eight cell identities, and a 120-cell lymph-node patch with seven identities. Each has local sensing, constrained cellular actions, finite resources and explicit clocks. Their biological mechanisms and recording formats are distinct.

What does “each cell is an AI” mean?

Each cell is modeled as an individual agent with its own state and local observations. In live experiments, Jev chooses among legal actions for eligible cells. Cells do not each train a separate model, and quiet cells do not make requests. The gut examples use a deterministic fixture policy; the lymph-node example contains actual recorded Jev responses and explicitly disclosed proposed fate criteria. This website only replays saved results.

What are the scientific limits?

This is an uncalibrated mechanistic demonstration, not a validated biological predictor. Choice probabilities are not biological event rates. IBD-like injury represents a limited innate inflammatory challenge, not a complete disease model.

Where are the rules and evidence?

Manual v3 documents the mechanisms, assumptions and primary references. It defines 56 action contracts; 22 physical handlers are implemented. Unsupported actions stay blocked.

Read the complete manual ↗

The people behind the project

Conceived and created by Helder Nakaya, leader of CSBL and founder of Hylix.app. Developed with ChatGPT and Claude as AI collaborators.

Project credits ↗