NATIVE WORKBENCH / PYTORCH + PYGAME
Morpho
Return to the
same living world.
Grow synthetic tissue, make an intervention, and keep the exact state that follows. Save the world, its neural update rule, and its random stream together.
Actual CUDA workbench capture · step 160 · untrained neural update rule with structured growth prior.
WHAT YOU CAN DO
- Pause, single-step, inspect seven state layers, and cut, burn, delete, feed, or poison an organism.
- Save and restore the complete session, then continue the same future.
- Resume genome evolution from a checkpoint and verify archived organisms against their original rollout.
Extract the release, install Python 3.11 or newer, and open a terminal in its folder. Start on CPU; use a compatible PyTorch CUDA build for your GPU.
python -m venv .venv
.venv/Scripts/python.exe -m pip install -e ".[dev]"
.venv/Scripts/python.exe scripts/run_viewer.py --config configs/tiny.yaml --device cpuF5 saves the session; F9 restores it. The README includes evolution, replay, and CPU/CUDA validation commands.
OBSERVED / CONTINUITY CHECK
The saved world
really resumes.
On the tested CPU and CUDA backends, restored sessions and evolution runs matched their continuous controls. All three archived organisms on each backend replayed exactly. Changing the recorded update rule made the negative control fail.
Inspect the replay report ↓PyTorch 2.9.0+cu130 · RTX 5060 · identical-source and same-backend checks. This is continuity evidence, not evidence of learned biology. Cross-version and cross-hardware replay is not promised.