EverMind-AI/Raven

The Harness of Harnesses • built for RSI: a trusted, persistent, self-evolving multi-agent ecosystem for all-domain collaboration.

What it solves

Raven tackles the complexity of coordinating multiple AI agents to complete large, multi-step projects. Instead of a single agent doing everything, Raven acts as a host that generates workflow graphs (DAGs) and orchestrates specialized agents for research, coding, design, and unattended automation. It also aims at recursive self-improvement, meaning the system can propose and adopt changes to its own planning and execution logic.

How it works

Raven splits its agent loop into four decoupled strategy modules: Memory (what a turn sees), Planning (how work is approached), Capability (which tools are available), and Action (what to do next and how to judge it). A separate component called the Curator rewrites these modules, installing verified improvements. Built-in agents include Raven-Research for deep research, Raven-Code for software development, Raven-Design for visual deliverables, and Raven-Oncall for long-running workflows. Raven can also connect third-party agents over ACP, CLI, or OpenAI-compatible APIs. It relies on EverOS for cross-session memory and context.

Who it’s for

Developers and teams who want an autonomous, multi-agent system to run complex projects end-to-end, such as building software, conducting research, creating design assets, or running unattended experiments. It also appeals to researchers interested in agent self-improvement, since Raven can evaluate and adopt its own harness changes. The README notes it is pre-alpha, so it suits early adopters comfortable with rapid interface changes.

Highlights

  • Orchestrates multiple built-in and third-party agents under one host, with DAG-based workflow planning.
  • Four built-in agents: Research, Code, Design, and Oncall, each claiming SOTA results on respective benchmarks.
  • Runtime self-evolution via a Curator that rewrites memory, planning, capability, and action modules, with verification before adoption.
  • Demonstrates full autonomous project delivery: a game, a recursive self-improvement research project, and a product launch, all produced by Raven with no human coding.
  • Can connect 13 third-party agents (Claude Code, Codex, etc.) through a unified interface.
  • Installable via a one-line script, Docker, or from source, with a web UI at localhost:18793.

Related

  • Dispatch
    Raven: The Harness of Harnesses for Recursive Self‑Improvement

    Raven is an open‑source, memory‑first multi‑agent platform that orchestrates built‑in and third‑party agents, supports runtime self‑evolution, and demonstrates end‑to‑end project delivery and recursive self‑improvement benchmarks.

  • Project
    8treenet/goraven

    GoRaven is a self‑hosted AI‑agent platform that gives each team member an isolated workspace where the agent can read/write files, run commands, call APIs, and use RAG‑based knowledge. It ships as a Docker image, supports any LLM API, offers a skill marketplace, plugin hooks, and admin‑controlled quotas, all under an Apache‑2.0 license.

  • Project
    AlphaAvatar/AlphaAvatar

    AlphaAvatar is a self-hostable framework for building proactive, multimodal personal AI assistants that can remember, plan, and execute tasks using a plugin-based architecture.

  • Project
    mikeyobrien/ralph-orchestrator

    A hat-based orchestration framework that keeps AI agents in an iterative loop with automated validation gates until a task is fully completed.

  • Project
    2FastLabs/agent-squad

    A flexible, lightweight framework for orchestrating multiple AI agents across Python, TypeScript, and Swift, featuring intelligent routing and context management.