README.md
What it is
The setup I work with every day. Claude Code and Claude Desktop run on several isolated accounts (personal and work), each with its own tools and permissions, driven by a single command: agents.
The pieces
- Isolated profiles: every account has its own configuration, MCP servers and tool guards.
- Local console: today, tasks, memory, connections and system status in one interface.
- Brain: a multi-user memory and task service, the same from terminal, desktop and phone. The service enforces the writing rules; search works by words and by meaning on local embedding models.
- MCP servers: some written by me (tasks, Google, Coolify, Syncthing), the rest official, one per account.
Why
Agents become reliable when the environment around them is designed: memory that doesn't rot, secrets in a vault and never in chat, explicit confirmation before every production write.
Status
A Tauri desktop app that updates itself on two channels (stable and beta). I'm on the 1.0 beta: I'm testing it on my own machines, then release candidate and 1.0 with the public repo.