NVIDIA unveils Agora Git memory for swarms
NVIDIA researchers unveiled Agora, an append-only shared memory architecture designed to coordinate decentralized multi-agent research swarms. By recording hypotheses, experiments, and results as immutable Git commits, Agora enabled 13 autonomous workers to coordinate across 1,703 contributions and close 62% of the gap to a trained model.
Git is proving to be the optimal decentralized coordination and consensus primitive for autonomous AI swarms, demonstrating that cumulative shared state yields vastly higher returns than brute-force parallel agent compute.
* Eliminates Swarm Redundancy: Traditional parallel agents burn compute repeating each other's experiments; Agora’s shared DAG ensures workers build on verified prior art.
* Verifiable Lineage: Modeling insights and verifications as Git commits guarantees reproducible provenance where any agent can check out and validate a claim.
* Monoculture Prevention: Built-in diversity-aware selection mechanics prevent the agent collective from premature convergence on early local optima.
* Validated Long-Horizon Autonomy: 13 autonomous workers collaborating for 12 days without human tasking proves decentralized multi-agent autoresearch is viable at scale.
DISCOVERED
1h ago
2026-09-17
PUBLISHED
1h ago
2026-09-17
RELEVANCE
AUTHOR
omarsar0
