agentglass CLI
One desktop window for Claude Code, Codex, Gemini CLI and OpenCode: terminals, pull requests, diffs and approvals
- 347 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The app inserts its PreToolUse hooks into global Claude Code settings and sees every tool call of every session
- Agents get a browser with your sign-ins, and the orchestrator and workers act as you with git and a terminal
- The phone pairs with the computer and can approve calls or read terminals at the granted access level
Install
- Works in
Manual install
python3 hooks/install_hooks.pyRun from a clone of the repository. Hooks go into the global ~/.claude/settings.json and a backup of the file is kept. For one project, add --project with a path.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The app inserts its PreToolUse hooks into global Claude Code settings and sees every tool call of every session
- Agents get a browser with your sign-ins, and the orchestrator and workers act as you with git and a terminal
- The phone pairs with the computer and can approve calls or read terminals at the granted access level
This is third-party code. Review the repository files before installing.
What it does
agentglass puts into one window everything a person faces when several coding agents work: a terminal per agent, pull request tabs, a diff editor and permission prompts. The app attaches to the tmux sessions where Claude Code, Codex, Gemini CLI and OpenCode already run and can start new ones: a task or a GitHub issue becomes a worktree, a branch and an agent, and several agents do not collide. The database is a SQLite file on your own disk, with no account and no cloud. The Lantern panel shows who needs you: red marks an agent stopped on a permission gate, amber one that finished and waits for input. A tool call you chose to gate is held until you answer, from the desk or the phone, and the queue sits on disk, so a crash cannot let it through silently. Agents get a browser panel they drive through a CLI and an MCP server, already signed in where you are signed in. Diffs show word-level changes grouped by agent, and pull request review combines checks, conversation and files in one form.
Who it is for. Developers who keep several Claude Code and Codex agents in parallel worktrees and drown in terminal windows, pull request tabs and permission prompts.
Good fit when
- Three agents edit different branches and you need to see at once which one stopped on a permission prompt
- Dangerous tool calls such as git push should wait for your decision, including from a phone
- An agent wrote a change and you want to review the word-level diff and accept only the parts you need
- GitHub issues are handed out to agents, each in its own branch and worktree
- An agent needs a browser with your sign-ins to check a page itself
Not a fit when
- One agent works in one folder: a terminal and the standard permission prompts are enough
- Agents run in the cloud or on a remote machine with your computer not nearby: the window is tied to one machine
- Hooks that edit the global ~/.claude/settings.json are forbidden by policy on this machine
- You only need token and cost counting: a separate usage monitor is simpler for that
Example request
Wire the agentglass hooks into Claude Code and show which of my sessions are waiting on a permission nowLimitations
- The macOS build is not signed, and Gatekeeper calls it damaged. One run of xattr -dr com.apple.quarantine /Applications/agentglass.app fixes that.
- The Windows build passes CI but has not been verified on real hardware. The Android app works only as a companion to a computer with remote access turned on.
- It needs git, Claude Code CLI and Python 3. The pull requests panel needs GitHub CLI with a login, containers need Docker, live panes need tmux.
- The orchestrator runs workers as you under a policy you set. It is a lock, not a sandbox, and SECURITY.md describes where it stops.
How to disable. bun run setup:undo removes the agentglass hooks from the global ~/.claude/settings.json, and Settings, Hooks in the app does the same. Then close and uninstall the app as usual. Agents in tmux keep running without it.
FAQ
How do I install agentglass?
Download the installer for your system from the releases page: AppImage or deb for Linux, dmg for macOS, exe for Windows. The server is already inside the app. To get live events and call gating, add the hooks with python3 hooks/install_hooks.py or the button in Settings.
Does agentglass replace Claude Code?
No. Agents stay the same CLIs on your machine and start with your login, and the app does not proxy them. If you close agentglass, agents in tmux keep running.
What leaves my machine?
By default nothing: data sits in a SQLite file, with no account and no telemetry. Only a webhook you configure and the diff hunks you send with Explain can leave, and both are off. The phone connects over your own network through a pairing code.
How do approvals from the phone work?
The Android app pairs with the computer by a scanned code and gets access at a chosen level: read, answer or full. Permission requests wait in a queue on disk, and you can answer from the computer or the phone.
What do I need installed first?
git, Claude Code CLI and Python 3 are required. The rest depends on features: tmux for live agent panes, GitHub CLI with a login for the pull requests panel, Docker for containers. Settings, Requirements checks all of it on the machine.
Discussion
No questions or reviews yet.
Related
CLI toolsA terminal shell for working with coding agents: embedded terminals, diffs, files, td tasks and worktrees side by side
agentplay turns local Claude Code, Codex and Cursor sessions into a file map and a replayable step timeline.
A desktop and browser workspace for parallel Claude Code, Codex and OpenCode sessions in separate git worktrees
Session monitor for Claude Code, Cursor, Codex and six more agents in a terminal, window or HTTP API, with chat cleanup and search
A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
A Rust coding agent with a native GUI, a terminal mode, an ACP mode for Zed and an MCP server
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| agentglass (this card) | no | 347 | |
| Agent Deck | no | 1,216 | a terminal session manager, without diffs, pull request review or phone approvals |
| Claude Squad | no | 8,738 | runs agents in tmux and worktrees in a terminal app, with no desktop window or permission queue |
README in short
The README explains the idea of one window that gathers agents, terminals, pull requests, diffs, a browser and approvals. It then lists capabilities, a platform table with honest status notes, installation and a section on data staying local. The page ends with an FAQ and a documentation table: INSTALL, WORKSPACE, CONFIG, CLEANUP and POLARIS.