为构建者与 Agent 打造的项目大脑。MindMux 是一款本地优先的 AI 工作台，拥有持续的项目上下文。讨论、提炼、分发——你的项目上下文会在会话、模型与 Agent 之间保持完整。
MindMux is a local-first AI workbench built around a project brain. It gives product and engineering teams one place to discuss ideas, distill stable knowledge, and dispatch execution without losing the reasoning that led there.
Unlike a general chat client, MindMux treats conversation as an interface, not the final output. The durable asset is the brain/: structured Markdown documents, knowledge pages, and execution context that remain usable across sessions, runtimes, and agents.
MindMux exists for teams that already know chat alone is not enough. Good discussions happen in chat, but durable projects need:
MindMux is designed around a simple loop:
MindMux keeps durable context in a local brain/ directory instead of trapping it inside a single vendor's chat history. The brain contains:
Sessions are first-class objects. They can be resumed, forked from a specific point, or spawned into a new line of work while staying connected to the same project brain.
MindMux separates the knowledge layer from the execution layer. Claude Code, Codex, and other runtimes can work against the same brain without forcing users to rebuild context every time the runtime changes.
MindMux can turn a settled discussion into a task, attach the relevant brain context, and dispatch it outward. The task system is intentionally lightweight: it is an execution bridge, not a project management suite.
The top-level unit users work in. Public-facing language should usually say "project", even though parts of the implementation use the term "workspace".
The durable knowledge layer stored as files. This is the long-lived memory of the project.
A discussion thread with its own runtime, expert, lineage, and message history.
The in-app discussion role that shapes how a session reasons and responds.
The execution pair that determines which model system runs the conversation or task.
The outbound execution unit used when discussion becomes actionable work.
