Mitosis run replay

Every step a model proposes passes a deterministic checkpoint before anything runs or reaches memory. Click agents, states and log entries to explore, or open the Lab ↓ to break the rules yourself.

0
transitions
0
actions blocked
0
outputs quarantined
0
agents terminated

Agents

click a cell to follow it
active dividing done apoptosis

Outcome

Cell cycle

click a state or arrow to jump to its next use

This step

Event log

following

    Memory

      Only an agent's own namespace is visible to it. Every change after birth went through an interceptor seal.

      Lab

      Runs the repository's actual Python package in your browser through Pyodide, not a JavaScript copy. Switch rules off, tune drift, write your own model output, and watch what changes.

      Engine not loaded · about 12 MB from jsDelivr on first use

      1 · Configure the interceptor

      Rules (uncheck to switch off)
        Drift policy
        Budgets

        2 · Propose an action

        Be the model. Write the JSON an agent would emit and the interceptor judges it as you type. If it's an allowed tool call, the tool really runs and its output is checked too.

        History so far (feeds the drift score)

        Load the engine, then edit the JSON or pick a preset.

        3 · Re-run the scenario

        Try an experiment, or edit what each agent's scripted model says and run the whole system with your settings. The result opens in the replay above as a new tab.

          Model script (per agent id)

          No live run yet.

          Generated by python visualize.py. The replay works offline; the Lab downloads the Python runtime when you load it.