Versions:
agtop is a process monitoring tool developed by Matt Brassey, designed specifically for observing AI coding agents such as Claude Code, OpenAI Codex, Aider, Google Gemini, and Block Goose. Currently at version 2.5.0 with eleven published versions in its history, the software occupies the system monitoring and developer tools category, presenting its data through a top-style terminal user interface built with the ratatui library. Its primary purpose is to give developers and operators real-time visibility into the resource consumption and activity of AI agents running on their machines. On Linux, agtop gathers process information by walking the /proc filesystem, while on Windows, macOS, and BSD systems it relies on the sysinfo crate for equivalent data. In addition to standard process metrics, the tool parses the on-disk session transcripts maintained by Claude Code, Codex, Goose, Aider, and Gemini, allowing it to correlate system-level activity with each agent's current behavior. For every detected agent, agtop reports CPU usage, resident set size, process status, the current tool or task in execution, in-flight subagents, cumulative token usage, estimated cost, context-window fill, loaded skills, and a live preview of the ongoing transcript. Use cases include tracking how multiple concurrent AI agents affect system resources, monitoring token expenditure and cost accumulation across sessions, and diagnosing what an agent is actively doing at any moment. The software also provides platform-specific capabilities: on Windows, writable file descriptor enumeration is performed using NtQuerySystemInformation combined with DuplicateHandle to surface which files each agent has open, and per-process IO byte counters are derived from sysinfo's disk_usage() function. By combining conventional process metrics with agent-specific transcript analysis in a single terminal dashboard, agtop serves developers who run AI-assisted coding workflows and need consolidated, up-to-date operational insight across heterogeneous agent ecosystems and operating systems.
Tags: