A distributed search for AGI

Intelligence has more than one path. We're searching all of them.

AI has converged on a single bet: scale transformers with backpropagation on ever-larger data and ever-larger compute. It works, but it treats one path as the only one, and it concentrates that path in whoever can afford the most hardware. SAGI starts from what we don't know: the architecture of general intelligence is still an open question. So instead of scaling one design, the network searches the space of possible learning systems, their architecture, their memory and the rules by which a mind changes as it learns, on compute contributed by everyone.

Every algorithm is an organism you can watch learn

Each candidate learning system is described by a genome and grown into a small living organism. You can watch it remember, adapt, recover from mistakes and respond when the rules of its world change. An abstract learning rule becomes something you can see, cultivate and care about.

A distributed search, running on everyone's hardware

Evolution Strategies are naturally distributable: each machine only exchanges random seeds and a single fitness score. That makes a worldwide network of laptops, GPUs and cloud nodes not just possible but efficient.

• Genome. A candidate mind is described as a compact genome.
• Organism. The genome is grown into an organism and placed in many environments.
• Your hardware. Your machine evaluates mutated offspring and returns their fitness.
• Selection. Evolution Strategies selects and mutates the population. Discoveries propagate.

Brute force scales one design. Evolution searches many, in parallel.

Any app can feed the swarm

The SAGI SDK lets developers plug any app into the network, contributing the human judgment, spare compute or proprietary signal the network cannot cheaply produce on its own. From a consumer app to a business's idle infrastructure, every contribution earns credits and ripples through the swarm in real time.

Every discovery is owned by the network that made it

A non-profit Foundation is the legal steward of everything SAGI discovers. It coordinates the search, verifies results and holds each breakthrough on behalf of the whole network, keeping it open and never enclosed by a single lab or nation. Sponsors can fund directed research and get a 90-day exclusivity window; after that every result opens to the entire network. Openness is constitutional.

Open doesn't mean unaccountable

Every bounty passes a two-stage review before it goes live: automated screening against a prohibited-category list, then an independent ethics committee with binding veto power.

• Prohibited by design. No autonomous weapons, mass surveillance or political disinformation.
• Independent oversight. A rotating committee of researchers, civil-society voices and legal experts.
• Community flagging. Anyone can pause a live bounty for review.

Open problems

Intelligence isn't one score. SAGI measures transfer to unseen tasks, adaptation speed, memory efficiency and compute cost, each with its own leaderboard. Anyone can post a bounty to point the network at an unsolved problem.

SAGI is a research project. Network tokens are utility rewards for contributed compute and verified contributions, not an investment or a promise of financial return.

Join the network →