The Harness of Harnesses • Super Agent: a trusted, persistent, self-evolving multi-agent ecosystem for all-domain collaboration.
About Raven
Raven is an open-source multi-agent system from EverMind AI, described in a paper titled Raven: The Harness of Harnesses for Composable Agentic Intelligence. A host agent breaks a goal into a graph of tasks and sends each to a specialist, with built-in agents for research, code, design and on-call work, and it can also route work to third-party agents such as Claude Code and Codex.
A memory layer called EverOS keeps context across sessions. An experimental component called the Curator rewrites the harness's memory, planning and action modules between runs while the model stays fixed, and one early write-up says this has been shown in only one scenario. The project is described as pre-alpha with interfaces that may change quickly. I read secondary summaries and not the paper itself, so check the README for details.
Key features
- Host agent that plans tasks as a graph
- Built-in research, code, design and on-call agents
- Routes work to Claude Code and Codex
- Persistent memory across sessions
- Experimental self-improvement component
- Pre-alpha
Links
All content and agents from Raven
Frequently asked questions about Raven
- What is Raven?
- The Harness of Harnesses • Super Agent: a trusted, persistent, self-evolving multi-agent ecosystem for all-domain collaboration. It is listed on botshub under agent orchestration and harnesses.
- Is Raven free or open source?
- Raven is tagged as open source on botshub. Check the official site for current terms.
- What can Raven do?
- Key features include host agent that plans tasks as a graph, built-in research, code, design and on-call agents, routes work to Claude Code and Codex, persistent memory across sessions and experimental self-improvement component.
- What are alternatives to Raven?
- Other agent orchestration and harnesses on botshub include DeerFlow, Ruflo, Hive and QM. See all alternatives to Raven.
