lazyagent watches session data from coding agents — Claude Code (CLI and Desktop), Cursor, Codex, Amp, pi, and OpenCode — and shows what each one is doing in real time. No modifications to any agent are needed; it’s purely observational.
Three interfaces ship in a single binary: a terminal UI, a macOS menu bar app, and an HTTP API. They share the same engine and can all run at once. Two maintenance subcommands, prune and compact, clean up old or oversized transcripts.
Getting Started
- Installation — Homebrew, Go, or build from source
- Quickstart — first launch, flags, and combining interfaces
Concepts
- How it works — the observational model and the shared core
- Supported agents — paths, prefixes, and per-agent quirks
- Activity states — the state machine behind the color-coded labels
- Session info — every field lazyagent surfaces, and where it comes from
Interfaces
- Terminal UI — the default, bubbletea-powered TUI
- macOS GUI — the detachable menu bar panel
- HTTP API — REST + Server-Sent Events, with the interactive playground
Usage
- CLI reference — every
lazyagentflag, with syntax and examples - Recipes — end-to-end walkthroughs for common workflows
Maintenance
- Prune old sessions — delete chat files by age or orphaned-project filter
- Compact session files — truncate bulky tool outputs and thinking blocks in place
- Search chat transcripts — full-text search with highlighted snippets and an incremental SQLite index
- Show rate-limit usage — on-demand 5-hour and weekly window snapshot for Claude Code and Codex, with a pace indicator
Reference
- Editor support — how
$VISUAL/$EDITORare resolved - Configuration —
~/.config/lazyagent/config.jsonfield by field - Architecture — module map of the codebase
- Development — build targets and dependencies
- Roadmap — shipped features and what’s next