agentglass

agentglass is an open-source AI agent observability tool that puts every AI coding agent running on your machine onto one screen, showing live tool calls, token usage, and cost side by side. It is built for developers who run multiple coding agents at once and lose track of what each one did across a dozen separate terminals. Instead of checking outcomes afterward, you watch each agent as a card that reports its model, current activity, cost, turns, and errors in real time. agentglass works across Claude Code, Codex, Gemini CLI, Kimi, and anything that emits OpenTelemetry, so it is not tied to a single provider. It is described as a workspace rather than just a dashboard: alongside monitoring, it bundles source control, diff review, pull requests, tasks, Docker, a terminal, chat, and a file browser behind keyboard shortcuts, and can hold a dangerous command until you approve it. The app runs locally, storing data in SQLite with loopback networking so nothing phones home, and is distributed as desktop installers for macOS, Linux, Windows, and Android. Built with Bun, SQLite, React, Vite, and Electron, agentglass is released under the MIT license, making it a self-hosted option for teams that want agent visibility without sending telemetry to a third party.

Best for developers running multiple terminal-based AI coding agents who want live, provider-agnostic visibility into tool calls, tokens, and cost on one screen.