Give all your AI agents one shared plaintext brain

What it does

MindOS is a local-first knowledge base that sits between you and every AI agent you use. It stores your notes, resumes, workflows, and project context as plain text on your machine, then exposes them to tools like Claude Code or Cursor through a built-in MCP server. The idea is to stop re-pasting context into every new chat and instead let agents pull from—and write back to—a single, persistent brain.

The interesting bit

Instead of letting each agent lock its memory in a proprietary black box, MindOS inverts the model: the knowledge base is the source of truth, agents are stateless clients, and you retain full read/write access to every file through a GUI. Every agent retrieval and reflection is logged as local plain text, so you can correct the “mind” when the AI hallucinates rather than debugging inside a closed model.

Key highlights

• Built-in MCP server plus YAML workflow engine; agents connect with zero config and execute multi-step routines from your notes.
• Auto-distills conversations into reusable SOPs and profiles, turning ephemeral chats into structured knowledge.
• Everything lives as local Markdown/plaintext with Git auto-sync, path sandboxing, and an Agent Inspector to audit tool calls.
• Cross-platform surface area: desktop app, Web UI, CLI, browser clipper, and mobile (though the README emphasizes the core package).
• Supports ACP/A2A protocols for agent-to-agent discovery and delegation, with Phase 1 (Agent Cards + JSON-RPC) currently live.

Caveats

• ACP/A2A agent-to-agent orchestration is explicitly marked as Phase 1, so inter-agent delegation is still early.
• Remote MCP setup requires manual JSON configuration, per the docs, so the “zero config” experience is mainly for local stdio transport.

Verdict

Worth exploring if you juggle multiple AI coding tools and want a single, ownable memory layer you can audit with a text editor. Skip it if you’re looking for a hosted, team-wide solution—MindOS is aggressively local-first and personal.