Agent Tree is a Luxen desktop workspace for running master and subagent sessions with shared context, live status, and one place to inspect the tree as work branches out.

It is built for multi-agent coding and research workflows where context needs to stay compact and reusable instead of being recopied into every chat.

Agent Tree screenshot

Agent Tree architecture

• Renders the live master, lead, and worker tree
• Persists shared context entries that downstream agents can reference
• Switches between Codex and Claude at runtime
• Shows per-agent lane, status, and recent updates in one desktop view
• Keeps the control surface local with no hosted backend

• electron/ owns runtime integration, filesystem access, and the shared context store
• ui/ renders the tree, lane filters, and selected session inspector
• shared/default-state.json provides a clean development state for local runs
• bin/agent-tree.js launches the packaged app

npm install agent-tree-viewer
npx agent-tree-viewer

npm run check
npm run pack:check

• The desktop app stores state in Electron's user-data directory
• Only active agents render in the tree
• Shared context entries stay intentionally compact and focus on findings, symbols, references, and outcomes

Agent Tree is a Luxen project. This public repo reflects implementation work by Ganesh Talluri and Ishaan Ranjan.

MIT