Skip to content

jessemaxh/GlanceTerm

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6,746 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

GlanceTerm logo

GlanceTerm

See every AI agent at a glance. Never miss the one that needs you.

The terminal for running multiple AI coding agents in parallelClaude Code, Codex, Gemini CLI, opencode.

License: MIT Platforms Latest release Built on Tabby

English · 简体中文


GlanceTerm sidebar — live status for every AI agent tab

Running 5+ Claude Code / Codex / Gemini / opencode sessions across tabs? Stop Cmd-Tabbing to figure out which one is done. GlanceTerm puts every tab in one live-status side panel — 🟢 working · 🔵 done & waiting · 🟠 needs permission — and one click jumps you to the tab that needs you.

✨ Highlights

  • 🟢 Live status per tab — working / done / needs-you, from each agent's own hook events (no polling, no screen-scraping, no false positives)
  • 🎯 Click to jump straight to the tab waiting on you
  • 🔄 Restart-safe — reopen and every agent resumes its exact prior session (claude --resume, codex resume, opencode --session)
  • 🤖 Multi-agent — Claude Code first-class & tested; Codex / Gemini CLI / opencode adapters too
  • 🌳 Worktree isolation (optional) — right-click → run an agent in its own git worktree + branch, so several agents work one project without clobbering each other; auto-cleans a clean one on close, with a manager panel for the rest
  • 🧩 Zero workflow change — keep typing claude; the hook self-installs on first launch
  • 📊 Token usage — in / cache / out per agent · session · project, with CSV export
  • 🛡️ Opt-in auto-approve (audit-logged, off by default) · 📸 screenshot-to-paste · split shell
┌────────────────┬──────────────────────────────────┐
│  AI TABS       │   you@host ~/work/api $          │
│ ● ai-backend   │   > what does this function do?  │
│   working      │   ⏺ Reading src/handler.ts…      │
│ ○ ai-frontend  │                                  │
│   ready  •     │                                  │
│ ◐ ai-tests     │                                  │
│   needs you    │                                  │
└────────────────┴──────────────────────────────────┘

Install

macOS (Apple Silicon) — Homebrew:

brew install --cask jessemaxh/glanceterm/glanceterm

…or grab the signed + notarized .dmg from the latest release.

Linux (x64) — from the latest release: .AppImage · .deb · .rpm · .pacman · .tar.gz.

Windows (x64) — …-setup-x64.exe or portable .zip from the latest release. ⚠️ Unsigned → SmartScreen "More info → Run anyway".

Every release builds and launch-smoke-tests all three platforms in CI automatically. macOS is Apple Silicon only; Linux/Windows are x64.

Agent support

Claude Code is first-class and validated in daily use. Codex's status detection is verified; Gemini CLI and opencode ship adapters not yet tested end-to-end.

Capability Claude Code Codex Gemini CLI opencode
Live status (working / done / needs-you) 🧪 🧪
Auto-approve permissions 🧪
Resume exact session on restart 🧪 🧪
Subagent + background-job badges

✅ tested · 🧪 implemented, untested with this agent · ❌ not available. Full per-event breakdown: docs/feature-matrix.md.

How it works

Each tab gets a unique GLANCETERM_TAB_ID at PTY spawn; claude (and others) inherit it. On first launch GlanceTerm installs a tiny hook in the agent's settings that writes a JSON status file to ~/.glanceterm/hooks/<tab-id> on each lifecycle event. The sidebar watches that directory and repaints each tab's status in real time — zero polling, zero screen-scraping, zero false positives.

⚠️ Auto-approve (opt-in, dangerous)

GlanceTerm can auto-answer allow to every Claude Code permission prompt — handy when babysitting many agents. Off by default. When on, Claude can run anything — including rm -rf or curl … | sh — without asking.

  • Every action is logged to ~/.glanceterm/auto-approve.log; toggle via the shield icon, or the ~/.glanceterm/auto-approve.flag kill-switch file.
  • Don't enable in your main repo, near credentials, or where sudo is passwordless. Use a container / scratch dir / disposable VM.

Build from source

Prereqs: Node 22, yarn.

cd glanceterm && yarn && npm run build
cd tabby-plugin-ai-sidebar && npm install && npm run build
cd .. && ./dev.sh        # launches the fork with remote debugging on :9222

Credits

Built on Tabby by Eugene Pankov. MIT-licensed, same as Tabby — see LICENSE and NOTICE. The SidebarProvider extension point added in the fork is intended to upstream.

About

The terminal for running multiple AI coding agents in parallel — Claude Code, Codex, Gemini CLI, opencode. A live status sidebar shows which session is working, done, or needs you, so you never miss the agent waiting on input. macOS, built on Tabby.

Topics

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Packages

 
 
 

Contributors