reddit · SideProject · 10-03 02:50 51 分 ← 返回列表

I built a shared brain for my coding agents (with memory, coordination and decision graphs)

I use Claude Code, Codex, Cursor and OpenCode. Every session starts from zero: none of them knows what the others did or what I decided last week. I was the memory, re-explaining the same things and copying context between windows. So I built MUON, a harness that runs your coding agents from one place, with a shared brain: One controller. Start, steer and stop every agent from one app/terminal. Each works in its own git worktree, so they don't step on each other. Shar

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：同时使用多个编码 Agent 时，每个会话从零开始，上下文、规则和历史决策无法共享，开发者只能自己充当记忆并在不同窗口间反复解释和复制。

目标用户：同时使用 Claude Code、Codex、Cursor、OpenCode 等多个编码 Agent 的开发者和小型工程团队。

证据：Every session starts from zero: none of them knows what the others did or what I decided last week. I was the memory, re-explaining the same things and copying context between windows. ... Free for individuals.

变现：个人免费 + 团队/企业付费：按席位收费，提供私有部署、审计日志、SSO、跨设备同步、更多 Agent 集成和权限管理；也可开源核心 + 云托管记忆/协调服务。

MVP：2 周内做一个本地优先的跨 Agent 共享记忆 MVP：基于 SQLite + MCP server，支持 Claude Code/Cursor/Codex 读取同一项目的决策、规则和上下文，按 git worktree 隔离，记忆需人工确认后写入，提供简单 CLI 查看和导出。

竞品：Cursor Memories、Claude Code memory/CLAUDE.md、Cline Memory Bank、Mem0、Letta、MCP memory servers、LangGraph、CrewAI

摘要：发帖人发布 MUON，为多个编码 Agent 提供共享大脑、协调图和决策账本，解决跨会话上下文丢失与重复解释问题。产品个人免费，目前仅支持 macOS Apple silicon，未提到付费或收入数据。痛点真实且方向热门，但付费意愿和外部需求尚未被验证。