NVIDIA's Agora uses Git as shared memory: 13 agents run unsupervised for 12 days
dair_ai · x · 2026-09-18
NVIDIA researchers published Agora (arXiv:2609.18094), tackling duplicate work when multiple coding agents research the same problem:
- Git as shared memory: every result, hypothesis, and verification is an immutable commit in an append-only DAG; parent edges show dependencies, and a derived index exposes the frontier and verification status
- A diversity-aware selection rule prevents collapse onto one leader
- First sustained run: 13 LLM workers, no assigned tasks and no central planner, ran nearly 12 days on weight transfer—initializing a 119.6M-param attention-SSM hybrid from 141 donor models with no matching dimensions, no training data, no gradient updates
- They published 1,703 contributions and drove the evaluator from 3.39 to 1.899 bits per byte, closing 62% of the gap to a trained GPT-2 124M
The winning recipe compresses donor next-token statistics into the target's embedding and output head, then adds a short-range context signal.
More from coding & agent
- Tencent unveils Hyra 2.0, a research agent that builds reusable scientific tools — TencentHunyuan · 2026-11-02
- General Reasoning explains how to train agent swarms with RL and latency pressure — rosstaylor90 · 2026-11-02
- "Coding isn't the bottleneck anymore — usage windows are," says founder Kyle Gawley — kylegawley · 2026-11-02
- Demo passed ≠ production-ready: 5 questions to answer before letting agents run your business — _kellhus · 2026-11-02
- An AI agent passed 100 test chats—still not ready to run the business — djgoel123 · 2026-11-02
- Agent given total creative freedom produces a 1:44 code-drawn animated short film — ivegotnoidea1 · 2026-11-02