Most multi-agent failures come from blind spots: you can't see running sessions, tool usage patterns, or how subagents spawn and interact. This project treats Claude Code as a first-class control plane, turning ephemeral agent runs into traceable, inspectable events so teams can observe, debug, and orchestrate agents in real time.
What Sets It Apart
-
Local-first observability pipeline: captures Claude Code session events via native hooks and stores them locally (SQLite) so telemetry and logs remain on-premise. So what? You get full session history and analytics without relying on external telemetry services.
-
Live control + UX: real-time dashboard with a Kanban-style status board, session inspector, and notifications that reflect live agent state via WebSockets. So what? Operators can intervene, pause, or spawn agents as they run, reducing time-to-debug for autonomous workflows.
-
Embedded MCP server & extension layer: ships a small MCP implementation and agent extension/skill layer to expose a tool catalog and enable programmatic interaction with the dashboard. So what? The monitoring UI can also act as a management endpoint for agent tooling and integrations.
-
Developer-friendly stack and integrations: built with Node.js/Express, React, TypeScript and SQLite, and provides hooks installation for Claude Code. So what? Teams can adapt or extend the dashboard with familiar web tooling and local-first constraints.
Who it's for and tradeoffs
Great fit if you run Claude Code agents locally or on private infrastructure and need visibility into sessions, tool calls, and subagent orchestration; if you want a self-hosted control plane that integrates with agent workflows and can be extended via MCP plugins.
Look elsewhere if you need a vendor-agnostic multi-LLM observability layer out of the box (this project is designed specifically around Claude Code hooks), or if you require enterprise-grade RBAC/compliance features that go beyond a local-first dashboard.
