// the find
hoangsonww/Claude-Code-Agent-Monitor
🚀 A real-time monitoring dashboard for Claude Code & Codex, built with SQLite3, Node.js, Express, React, Vite, TailwindCSS, & WebSockets. It tracks sessions, agent activity, tool usage, and subagent orchestration, providing live analytics, a Kanban status board, status notifications, a cute buddy, & an interactive web UI/MacOS/Windows native app.
A local dashboard that tails Claude Code, Cursor, and Codex session logs via hooks and renders them as a live fleet view — sessions, token/cost tracking, tool-call timelines, and orchestration graphs for parent/child and subagent runs. It's aimed at people running several AI coding agents at once who want one place to see what they actually did, not just what they were told to do.
Local-first by default (binds to 127.0.0.1, no cloud call needed to see your own data) is the right default for a tool reading your agent transcripts. SQLite via better-sqlite3 in WAL mode is a sane choice for a single-writer local app instead of standing up a real database. It actually ships tests (Vitest, RTL) and CI rather than being a demo repo with a flashy README. The CLI plus MCP server means you can query sessions/costs/workflows programmatically instead of only through the web UI, which is a real feature, not padding.
The README has roughly 50 badges spanning AWS, GCP, Azure, Oracle Cloud, Kubernetes, Helm, Terraform, Prometheus, and Grafana for a tool whose core job is reading local JSONL transcripts into SQLite and pushing them over a WebSocket — that's a lot of enterprise-infra signaling for what should be a lightweight devtool. Claiming 14 plugins, 66 plugin skills, 18 subagents, 34 commands, and 77 repo skills across two different providers' plugin formats is a huge surface to keep working as Claude Code and Codex each evolve their own skill/plugin schemas independently; this is the kind of thing that quietly breaks on provider updates. There's an Electron desktop app with native installers for both OSes, plus a whole separate k8s/Terraform/Helm deployment story, on top of five translated READMEs — none of that is wrong to have, but it's a lot of packaging surrounding what's fundamentally a single-machine monitoring tool, and it reads like scope spread rather than depth on the core dashboard. No mention anywhere of log/transcript retention or SQLite write contention if you're running several concurrent, chatty agent sessions — that's exactly the failure mode a monitoring tool for monitoring tools should have an answer for.