lazychat

One terminal for all your AI coding agents.
Run Claude Code and Codex sessions side by side, see at a glance which one needs you,
and keep Git and a shell for every project one key away.

Website · Install · Quick start · Features · Sponsor Lazy ♥ · Contributing

Install

You need a Mac (Apple silicon or Intel, macOS 12 or newer) and an AI coding agent you are logged in to: Claude Code or Codex.

Homebrew

brew install perpeer/tap/lazychat

Homebrew builds lazychat on your Mac, along with Lazy's menu bar app, so macOS opens both without a warning. Then run lazychat; lazychat doctor says what else is ready. Lazy starts with lazychat; brew services start lazychat keeps it in the menu bar from every login.

To update:

brew update && brew upgrade lazychat

lazychat comes from our own tap for now. Homebrew's own list takes it at 225 stars, and then it is just brew install lazychat. If it saves you a few tabs, star it: every star helps lazychat reach more people.

From the source

git clone https://github.com/perpeer/lazychat
cd lazychat
./install.sh

It checks your Mac, says what is missing, and builds ~/.local/bin/lazychat and Lazy's menu bar app. To update, run git pull && ./install.sh in the same folder.

Updates

When a new release is out, a green new box shows on the left rail over Settings (and ↑ 1.0.5 beside the version at the bottom right). Click it, or press U: a small window shows the Homebrew command above, and c copies it. With Homebrew, Enter upgrades right there in the background, showing the step it is on, then offers to restart lazychat on the new version — running sessions stop and stay in the list. If the upgrade fails it says so and shows the command to run by hand. Homebrew sees a release within the hour it comes out. From the source, Enter runs it in a new shell in the Terminal tab instead. If Lazy runs as a service, brew services restart lazychat brings the new menu bar app too.

To remove lazychat, run brew uninstall lazychat, or ./uninstall.sh for a source install. Your settings and workspaces stay (./uninstall.sh --purge takes them too); your projects and the agents' own files are never touched.

Quick start

1. Run lazychat and name a workspace, the set of projects you work on.
2. Press o and pick a project folder.
3. Press n to start Claude Code or Codex in it, Enter to type to it, and ctrl+q to come back.
4. Tab moves between the tabs, and ? lists every key where you are.

lazychat doctor tells you which agents are ready, and why not.

Features

Chat: your agents, side by side

Every session sits under its project with its terminal next to it. You can see which one works, which one waits for you and how long the last prompt took. n starts one, r resumes an old one, and w lets you draft the next prompt while the agent is still busy. i, or a click on Lazy while several wait, opens the inbox: the sessions asking you something or finished, Enter opens one. Press 3 for the details. At the top is the prompt's flow, drawn like git log --graph: each tool it called in order with the file it touched or the command it ran, a subagent as a branch that forks off and joins back when it returns, a question to you, and what each step added to the context. Beside it is the session's context, drawn like Claude Code's /context: how full the window is, what fills it, which tool's results took the most, and how many more prompts fit at this pace. At the bottom are your last ten prompts as a table with their state, duration, tokens and API cost. Lazy says hello when lazychat opens, and plays a sound when a session starts on a prompt, asks you something, finishes or fails; you can turn the sounds off in Settings.

┌ [1] projects ──────────┐┌ [2] session │ [3] details ───────────┐
│ garden-shed            ││ ❯ paint the shed blue                │
│ main                   ││                                      │
│   │                    ││ ● Starting with the north wall.      │
│   ├─ ● ivy             ││   Reading paint.go…                  │
│   │    claude · 42s    ││                                      │
│   │                    ││ > ▌                                  │
│   └─ ○ oak             ││                                      │
│        codex · 3m 18s  ││                                      │
└────────────────────────┘└──────────────────────────────────────┘
session: (enter) continue · (n) new · (r) resume · (w) draft · (?) help

Git: stage, diff and commit without leaving

Changes as a folder tree, a diff you can walk line by line with the code in its language's colours — v selects lines, space stages just those — your last commits and the repository's worktrees. Suggest asks the agent for a commit message written from what you staged.

┌ [1] projects ──┐┌ [2] Unstaged · 1 ────┐┌ [5] diff ──────────────────┐
│ garden-shed    ││ M paint.go           ││ @@ -3,2 +3,2 @@            │
│ main ↑1        │└──────────────────────┘│ - color := "red"           │
│                │┌ [3] Staged · 1 ──────┐│ + color := "blue"          │
│ blue-door      ││ + door.go            ││                            │
│ worktree       │└──────────────────────┘└────────────────────────────┘
│                │┌ [4] commits ─────────┐┌ [6] commit ────────────────┐
│                ││ ↑ a1b2c3d first coat ││ Paint the door blue        │
│                ││   9f8e7d6 sand it    ││────────────────────────────│
│                ││                      ││      [ Suggest ] [ Commit ]│
│                ││                      ││                            │
└────────────────┘└──────────────────────┘└────────────────────────────┘
branch: (c) commit · (p) pull · (shift+p) push · (b) branches · (w) worktrees

Terminal: a shell in every project

Plain shells in your projects' folders, for the things the agents don't do.

┌ [1] projects ──────┐┌ [2] sh 1 ──────────────────────────────┐
│ garden-shed        ││ $ go test ./...                        │
│ main               ││ ok   shed/paint   0.4s                 │
│   │                ││ $ ▌                                    │
│  ├─ ● sh 1         ││                                        │
│  │                 ││                                        │
│  └─ ○ server       ││                                        │
└────────────────────┘└────────────────────────────────────────┘
terminal: (enter) continue · (n) new · (e) rename · (v) copy · (d) close

Settings holds the theme, which tabs show and which agent writes your commit messages. Every key and screen is in the reference.

Sponsor Lazy

Lazy watches your agents all day and has never asked for anything. lazychat is free and stays free; if it saves you time, keep Lazy going:

☕ $3 a coffee · 🍩 $10 a snack · 🛋️ $50 a comfy cushion · 🚀 $100+ super sponsor

Inspiration

lazychat follows lazygit and lazydocker by Jesse Duffield: terminal UIs where everything is one key away. lazychat does the same for AI coding agents.

License

AGPL-3.0, see LICENSE. lazychat is free to use, change and share; a changed copy that is shipped, sold or run as a service has to stay open source too, and keeps the credit in NOTICE.

Contributing

Bug reports, ideas and pull requests are welcome: open an issue and tell us which agent lazychat should support next. CONTRIBUTING.md explains how the code fits together and how it is tested.