Abtop Open source
htop-style live dashboard for your Claude Code, Codex and OpenCode sessions
Abtop ecosystem
What Abtop can load, and the skills, subagents and MCP servers in this directory that work with it.
- – Agent Skills
- – Subagents
- – MCP
- – AGENTS.md
- – Plugins
- – Hooks
- – Rules
🧩 Agent skills for Abtop
Nothing listed yet.
🔌 MCP servers for Abtop
Nothing listed yet.
What is Abtop?
What it does
Abtop is a terminal dashboard, styled after htop, purpose-built to monitor AI coding agent sessions rather than system processes. Instead of CPU/RAM per process, it shows per-session token usage, context-window percentage, rate limits, git status, running child processes, open ports, and current task, refreshed live while you work.
What is inside
This isn't a tool the agent calls — it's an operator-facing dashboard you run alongside your agent to watch what it's doing. Panels include: token/context usage, rate-limit countdowns, session status, child process/port trees, and git status, with 12 built-in color themes (including colorblind-friendly options).
Works with
- Claude Code — full support (token usage, context %, rate limits, tasks)
- Codex CLI — supported, minus context window %, current task, rate limit and memory status
- OpenCode — limited support, and requires a local sqlite3 database
How to install or connect
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/graykode/abtop/releases/latest/download/abtop-installer.sh | sh
Also installable via cargo install abtop, a Windows PowerShell installer, or pre-built binaries from GitHub Releases. A 120x40 terminal is recommended (minimum 80x24).
Maintenance and safety
MIT licensed, written in Rust, with roughly 3.6k GitHub stars and active development. The README states it is "all read-only. No API keys. No auth," and doesn't transmit data over the network — though its JSON session snapshots can contain project context and should be handled as private. Because it never writes to your agent sessions or touches your source code, the main safety consideration is simply keeping any exported snapshot files private, the same way you would treat a screenshot of a terminal.
Who should use it
Anyone running multiple long-lived Claude Code, Codex or OpenCode sessions at once who wants a single always-on view of token burn, rate limits and session health instead of tabbing between terminals. It's a companion utility, not a coding agent itself — pair it with one of the tools it monitors, and expect the most complete picture when that tool is Claude Code.
Source code: graykode/abtop ★ 3.7k · +41 this week · MIT · updated 2026-09-14
Pros
- Read-only, no API keys or auth required, per its own README
- Tracks token usage, context %, rate limits and git status live across multiple agent sessions
- 12 themes including colorblind-friendly options; MIT licensed with ~3.6k stars and active development
Cons
- It's a monitoring companion, not a coding agent itself, so it's a loose fit for an 'app' directory built around agents
- Codex CLI and especially OpenCode support are partial (missing context %, current task, rate limit, or requiring local sqlite3 access)
Abtop pricing
Open source. Prices change often, so confirm on the official pricing page. See every vibe coding app's pricing side by side.
Abtop alternatives
All alternatives →Claude Code Paid
Anthropic's agentic coding tool for the terminal, IDE, desktop and web
🔥 Claude Pro billed annually: $17/month instead of $20OpenAI Codex Freemium
OpenAI's coding agent for the terminal, IDE and cloud, included with ChatGPT plans
OpenCode Open source
Open-source terminal coding agent that works with any model provider
Amp Freemium
Frontier coding agent for the terminal and editor, from the makers of Sourcegraph
Orca Open source
An ADE for running fleets of coding agents in parallel worktrees, from desktop or mobile
Qwen Code Open source
Open-source AI coding agent from Alibaba's Qwen team, for terminal, IDE and desktop
Compare Abtop
Verdict
Abtop scores 7.0/10 in our terminal coding agents ranking. Read-only, no API keys or auth required, per its own README.