Contribute Compute to REE

Donate spare GPU or CPU time to a public AI safety research project

🔍

Fully open source — read the code before running anything. Every experiment script that will run on your machine is at github.com/Latent-Fields/ree-v3/experiments. No surprises.

📄

Results only. Your machine trains small neural networks and pushes the result JSON to a public GitHub repo. No other data leaves your machine. You can inspect every output file before it's committed.

🔑

SSH key is read-only and scoped. The Docker container mounts your key read-only and uses it for one thing: git push to GitHub. It cannot access your filesystem, other keys, or any other service.

⚡

Low priority — stop any time. The runner runs at below-normal CPU priority. docker compose down stops and removes it completely. No background services, no startup entries, no persistence after removal.

👤

What this is. REE is a cognitive architecture research project exploring how AI agents can reason about ethical cost across time — not a product, not a company. Run by Daniel Golden, open Apache 2.0. Experiment results feed a public claims registry.

My platform:

🚀 Get Started

1

Install Python 3.12 and Git if not already installed. Then clone:

git clone https://github.com/Latent-Fields/REE_assembly.git
Both repos are public — no account needed to clone. You can browse experiment scripts at ree-v3/experiments/
2

Run the setup script — it installs PyTorch, configures git, registers your machine, and prints the runner command:

cd REE_assembly py -3.12 setup_contributor.py
The script guides you through each step interactively. Takes ~5 minutes (mostly PyTorch download). Prefer Docker instead? See Alternative Setups below.
✓

The runner starts automatically inside the container. After ~30 seconds:

http://localhost:8000/explorer
You should see runner status Running and the current experiment name. That's it — experiments run in the background, results push to GitHub automatically.
To share results with the project — the runner needs push access to GitHub. Request write access to the Latent-Fields org (usually approved within 24 hours). While you wait, the runner works fine locally — it will retry pushes automatically once access is granted.