Huge release from EverMind AI.

Raven is an open-source multi-agent system that builds a separate harness for each model and domain, evolves those harnesses from their failures, and coordinates them through a host agent.

The host agent splits a goal into a task graph and sends each part to a model-harness pair for research, code, design, or on-call work.

Harness components are diagnosed, mutated, and recombined, and a candidate replaces the current harness only after passing a statistical check.

With DeepSeek-V4-Flash, the research harness scores 69.3% on BrowseComp and 60.0% on Humanity's Last Exam, against at most 62.4% and 43.3% for two other harnesses on the same model.

Its curated skill library raises SkillsBench Pass@1 from 9.2% with no skills to 22.6%.

Paper: https://arxiv.org/abs/2609.33439

Chat with Paper: https://academy.dair.ai/papers/raven-the-harness-of-harnesses-for-composable-agentic-intelligence-2609.33439