EverMind AI has released Raven, a modular multi-agent orchestration framework designed to support recursive self-improvement (RSI). Raven acts as a host agent that integrates built-in and third-party agents to execute complex tasks through a modular architecture.
Product LaunchesEverMind AIRaven
EverMind AI Releases Raven, a Modular Multi-Agent Orchestration Framework for RSI
The framework includes four built-in agents: Raven-Research for technical analysis and literature reviews, Raven-Code for agentic software development, Raven-Design for visual asset creation, and Raven-Oncall for unattended workflow automation. These agents can be orchestrated to complete end-to-end projects, such as developing playable games or running large-scale training experiments.
A key component of the ecosystem is the Raven Evolver, a separate tool that treats Raven as a library to evaluate and propose improvements to the orchestration harness itself. Raven's architecture is composed of four decoupled modules: Memory, Planning, Capability, and Action, managed by a "Curator" that iteratively refines agent behavior and tools based on validation results.
Raven supports connections to third-party agents via ACP, CLI, or OpenAI-compatible APIs and offers a web-based user interface for managing multi-agent collaboration and workspace tasks. The project is currently in a pre-alpha stage and is available as an open-source tool.
Sources
- Show HN: Raven – The harness of harnesses, built for RSI （Hacker News Frontpage, 2026-09-29）
- Documentation