abtop Agent Session Monitor
A terminal dashboard for local AI coding sessions, tokens, context, limits, and ports.
Per-dimension scores and reasoning
Evidence shows: project claims read-only, no API keys, no auth, reads local files and process metadata, minimal permissions. But lacks user confirmation mechanisms (e.g., before killing processes), data flow transparency is described but not deep, sensitive data handling has privacy notes but JSON snapshots may contain sensitive content, dependency security not audited, external effects include kill and port operations without confirmation, rollback mechanism not mentioned, source attribution only author info. Deductions: user confirmation, dependency security, rollback, source attribution insufficient.
Evidence shows: README and Cargo.toml consistent, dependency list clear, error handling has hints (e.g., sqlite3 missing warning). But no detailed failure scenarios, dependency availability not verified. Deductions: failure messages not detailed enough.
Evidence shows: clear target audience (multi-agent developers), multi-platform support, clear capability boundaries (which agents, which features), clear trigger conditions (command arguments), environment fit documented (Windows no load average). Deductions: none.
Evidence shows: README structure clear, install instructions detailed, naming stable (abtop), examples and FAQ (partial), known limitations mentioned (e.g., OpenCode lacks some features), MIT license clear, version number exists but no changelog, maintenance responsibility not clear. Deductions: versioning changelog missing, maintenance responsibility unclear.
Evidence shows: output usable (TUI, JSON snapshot), high marginal value (solves multi-agent monitoring pain), cost-benefit reasonable (free open source). Deductions: none.
Evidence shows: README claims partially supported by code (e.g., supported features), but no test results or independent verification, facts and inferences not clearly separated. Deductions: claim traceability insufficient, cross-source corroboration insufficient, fact-inference separation insufficient.
- Kill process and port operations lack user confirmation, may cause accidental kills.
- JSON snapshots may contain sensitive project context, handle with care.
- Dependency security not audited, recommend checking for vulnerabilities.
- Version changelog missing, upgrade risks unclear.
What does this agent do, and when should you use it?
abtop is a Rust terminal monitor that presents local AI coding sessions through a TUI, one-shot text output, and JSON snapshots. It discovers Claude Code, Codex CLI, and OpenCode sessions from local process and file state, including multiple Claude configuration profiles. The interface surfaces token usage, context-window percentage, status, Git state, child processes, listening ports, and selected session-specific data, with configurable themes, panels, and language. It runs as a CLI and as a library: consumers can use App to collect state and serialize Snapshot data without requiring API keys or authentication.
abtop discovers Claude Code, Codex CLI, and OpenCode sessions from local process state, files, and open-file metadata; for OpenCode, it reads the local opencode.db through sqlite3. It aggregates token counts, context-window percentage, rate limits, current task, Git state, child processes, and port metadata for its TUI. abtop --once prints one snapshot, while abtop --json produces a one-shot JSON snapshot for scripts. Library consumers create an App, call App::tick_no_summaries(), and call App::to_snapshot(interval_ms) for a serializable Snapshot; that refresh path never runs claude --print.
- A developer running Claude Code in several projects can compare session state, token use, and context-window pressure from one terminal screen.
- A Codex CLI user investigating a forgotten development server can inspect session child processes and open ports.
- An OpenCode user with sqlite3 installed can view local OpenCode sessions alongside Claude Code and Codex CLI sessions.
- A team building a local dashboard can consume a one-shot state export through abtop --json or the Snapshot library API.
- A tmux, cmux, or macOS iTerm2 user managing several panes can select a session and press Enter to jump to its terminal.
What are this agent's strengths and limitations?
- Monitors Claude Code, Codex CLI, and OpenCode together, with local discovery that supports multiple active Claude profiles.
- Offers both --json and the reusable App/Snapshot library interface for local scripts or dashboards.
- Reports session-level token, context, limit, child-process, and port information rather than only host metrics.
- Keeps mouse capture off by default for terminal selection and copy, while providing 12 built-in themes, including four colorblind-friendly options.
- OpenCode session discovery depends on sqlite3, so it cannot be monitored without that CLI.
- Generating session summaries uses claude --print, which makes its own API call and affects Claude quota.
- OpenCode lacks context-window percentage, current-task, and rate-limit data; Codex CLI lacks subagent and memory-status data.
- The recommended terminal size is 120x40; panels hide below 80x24, and Windows reports LOAD as 0.
How do you install or deploy this agent?
On macOS or Linux, run curl --proto '=https' --tlsv1.2 -LsSf https://github.com/graykode/abtop/releases/latest/download/abtop-installer.sh | sh, or run cargo install abtop. On Windows, run powershell -c "irm https://github.com/graykode/abtop/releases/latest/download/abtop-installer.ps1 | iex"; cargo install abtop also works from a terminal with Git in PATH. OpenCode monitoring requires sqlite3; on Windows, install it with winget install SQLite.SQLite. Start the first working session with abtop.
How do you use this agent?
Run abtop for the interactive TUI, abtop --once for a single printed snapshot, or abtop --json for a one-shot JSON snapshot. abtop --setup installs the rate-limit collection hook; abtop --theme dracula selects a theme; and abtop --mouse enables mouse navigation. Configure theme, hidden_agents, claude_config_dirs, and language in ~/.config/abtop/config.toml.
How does this agent compare with similar options?
The README positions abtop as “like btop” for AI coding agents. In that framing, abtop is specialized for discovering local Claude Code, Codex CLI, and OpenCode sessions and presenting session token, context, rate-limit, process, and port data.