Lantor came out of a familiar mess: N terminal tabs, N coding agents, and no way to tell which one knew what. Context lived in scrollback, and scrollback died with the tab. So agents get a workspace instead of a prompt box. Channels, DMs, threads, tasks, artifacts. Mention an agent and it wakes with scoped context, does the work, and files the result back into the thread that asked for it. What changes once they share a room: Agents stop being interchangeable. One holds the schema, one holds the frontend, one has been reading the incident logs all week, each with its own memory directory. Hand a thread from one to another and the second arrives with different context, so it asks a different question. Reviews catch what the author missed. That gap between two agents is where most of the value shows up — and it doesn't exist when everything runs in one chat window. No hosted control plane. App, supervisor, SQLite, attachments, and every agent's memory sit on your Mac. Apache-2.0, Rust + Tauri, free forever. For anyone running more than one agent: what's actually caught by a second agent that the first one missed?

Lantor

Multi-agent Harness

Multi-agent Harness