Claude Code Multi-Agent Observability

Use it when

• Multiple Claude Code agents or subagents run in parallel and you cannot tell which agent issued which tool call or when a task handoff happened
• You want a persistent, filterable timeline of every hook event across several projects and sessions instead of scrolling terminal output, including tool failures and permission requests surfaced as they occur

What it solves

Not the fit when

• Production LLM application tracing and cost analytics (use an LLM observability platform)
• Evaluating model output quality or running regression evals
• Monitoring non-Claude-Code agent frameworks
• Exposing the server publicly: configured hooks persist prompts, event payloads, and transcript material in SQLite, so keep it on a trusted private network
• no-regression-signal
• workflow-retry-visibility

Install

Clone the repo, install Bun and Astral uv, run ./scripts/start-system.sh (or 'just start'); then cp -R .claude /path/to/your/project/ and set --source-app YOUR_PROJECT_NAME in .claude/settings.json for each project you want to monitor

Invoke

Open http://localhost:5173 while Claude Code runs; hook scripts POST all 12 lifecycle events (PreToolUse, PostToolUse, SubagentStart/Stop, SessionStart/End, etc.) to the Bun server on port 4000 and the Vue dashboard streams them over WebSocket with filtering and a live pulse chart

Alternatives

No reviewed alternatives recorded yet.