running ai agents
Give every agent its own pane
Tile a reviewer next to a builder, keep an eye on both, and jump into the files they touch with one click. Claude Code, Codex, or anything with a CLI.
→ This is what Tilzio is built for.
free · open source · native macOS · fully local
The terminal built for AI agents.
Run your AI coding agents side by side in a native macOS workspace — free, fully local, open source. Keep Claude Code and Codex usage in view, and jump straight into the files they touch.
Native macOS · no account, no cloud, no telemetry
// the usual
Five agent sessions across three windows, no idea which one is waiting — and Claude Code hits its 5-hour limit halfway through a refactor.
// tilzio
Every agent in its own pane, the finished one flags itself, and the limits are on screen before you hit them.
why tilzio Everything an agent workflow needs.
Everything an agent workflow needs.
Nothing it doesn't.
Tilzio stays out of the way: you bring your own agents, it gives them a workspace — and keeps you informed.
Run your own agents, side by side
Tile Claude Code, Codex, or any CLI agent across panes and spaces. No config file. Agent-agnostic by design.
pane 1 ✻ claude · refactoring auth…
pane 2 ● codex · running vitest…
pane 3 $ idle — yours
Usage & limits, in view
Claude Code and Codex spend, 5-hour window and weekly cap — on screen before you hit the wall.
5h 62%
wk 41%
cdx 12%
Click a path, land on the line
Built-in CodeMirror editor opens agent-printed paths at the exact line.
✔ done — see src/auth/session.go:42
Know when an agent needs you
The finished agent flags itself: a breathing pane highlight and a badge on its tab.
● api — claude · done
Free, local, open source
$0 — no tiers, no metering0 network calls home100% open — host + plugins
Native Go + Wails
One small native binary, fast startup, light on memory.
use cases
Built for agents — a serious terminal either way.
coming from tmux
The panes you love, as a real GUI
Drag-drop tiling, saved spaces, zero .tmux.conf — and Tilzio restores your layout and scrollback when you reopen it.
→ Local AI-agent work on your Mac is where it shines.
ops console
All your SSH sessions, one window
Keep sessions to your boxes tiled in one window, with named spaces per environment and a layout that comes back when you reopen it.
→ And when an agent run finishes in the next pane, the bell flags it.
grid consoles
Many sessions, one action
Open a grid of consoles at once — logs, builds, watchers, REPLs — instead of a pile of windows and tabs.
→ Agents slot into the same grid as everything else.
plain multiplexer
Also a serious multiplexer
No agents required: GUI tiling, spaces, a built-in editor and a plugin system on their own.
→ The AI features are there when you want them.
how tilzio compares
How Tilzio stacks up.
# A quick side-by-side across the things that matter for running agents in a terminal. Tilzio pinned on the left.
Yes Partial No
| Axis | Tilzio | Warp | iTerm2 | Ghostty | tmux | Zellij | Apple Terminal |
|---|---|---|---|---|---|---|---|
| Free & open source | Yes host + plugins | Partial client AGPL; AI backend closed | Yes GPL-2.0+ | Yes MIT | Yes ISC | Yes MIT | Partial free, closed-source |
| Native (not Electron) | Yes Go + Wails | Yes Rust, GPU | Yes Cocoa | Yes native + GPU | Yes C | Yes Rust TUI | Yes Cocoa |
| Local-first (no account/cloud/telemetry) | Yes | Partial AI needs cloud | Yes mostly | Yes | Yes | Yes | Yes |
| GUI tiling + spaces (no config) | Yes drag-drop, spaces, grid | Yes optional TOML | Partial config-heavy, no spaces | Partial no spaces/drag-tile | Partial text UI, .conf | Partial TUI, keyboard | Partial tabs + 1 split |
| Run multiple AI agents side by side | Yes agent-agnostic | Yes built-in orchestration | Yes per pane | Yes per split | Yes per pane | Yes per pane | Partial tabs only |
| Agent-finished alerts | Yes pane highlight + tab badge | Yes native | Yes native | Yes native OSC 9/777 | Partial visual bell | No needs plugin | Partial bell only |
| AI usage & limits visibility | Yes Usage Watcher: Claude + Codex | Partial own credits | No | No | No | No | No |
| Built-in editor + clickable paths | Yes CodeMirror, open-at-line | Partial no file editor | Partial no editor | No | No | No | No |
| Plugin system | Yes sandboxed slots | Partial MCP, no UI store | Partial Python API | No | Partial TPM | Yes WASM | No |
Agent orchestrators — a different category
# These run or orchestrate agents in isolated git worktrees, or are full IDEs. Tilzio is a terminal you run agents in, not a worktree orchestrator.
- Conductor — native Mac app; runs multiple Claude Code / Codex / Cursor agents in parallel, each in an isolated git worktree with diff and review-to-PR. Closed-source, account-bound.
- Crystal — open-source desktop app for parallel Claude / Codex sessions in worktrees.
- Vibe Kanban — open-source local web kanban orchestrating many agents in worktrees.
- Cursor — AI-native IDE (VS Code fork) with integrated terminal and parallel agents; account/cloud, credit-based.
- VS Code integrated terminal — xterm.js terminal inside the IDE; Copilot agent mode.
Switching from…
from tmux
Keep running each agent in its own pane — now as a real macOS GUI: drag-drop tiling, saved spaces, Gruvbox theming, zero .tmux.conf, and your layout and scrollback restored when you reopen. Plus a built-in editor with clickable paths, bell alerts, and a Usage Watcher for Claude Code + Codex limits.
from iTerm2
Tilzio trades config sprawl for an AI-agent-first GUI: drag-drop spaces, a built-in CodeMirror editor that opens terminal paths at the exact line, and a Usage Watcher showing Claude Code + Codex spend and 5h/weekly limits — all fully local and open source.
from Warp
If you liked running agents but not the cloud tradeoff, Tilzio is fully local: no account, no credit metering, no telemetry, and genuinely open source (host + plugins). Your AI value comes from agents you run yourself in panes, plus a local Usage Watcher with Claude Code + Codex limits.
from Ghostty
Tilzio adds what a pure emulator doesn't: drag-drop tiling spaces, a built-in editor with click-to-open-at-line, a plugin system, and AI usage/limits visibility for Claude Code + Codex — the workspace around your agents, not just the emulator.
plugins
The Usage Watcher is just a plugin.
Your limits, on screen — not in your face
Claude Code and Codex spend, the 5-hour window and the weekly cap live in the status bar. Fully local: it reads your own usage, no account required.
It's a regular Tilzio plugin, not a built-in — the same sandboxed API is open to yours.
slots: activity bar · status bar · bottom panel · right column · install from zip or URL
Give your agents a home.
Requires macOS 13+ · free · open source (MIT)
$ brew install --cask tilzio
# open source, top to bottom — the host and every plugin. Issues and PRs welcome.
faq
Straight answers.
Does Tilzio let me know when Claude Code or Codex finishes or needs input?
Yes. When an agent rings the terminal bell, Tilzio highlights the pane and badges the tab and space, so you spot it at a glance.
Can I run multiple AI coding agents side by side?
Yes. Each pane is a full terminal, so you can tile Claude Code, Codex, and any other CLI agents next to each other across panes and spaces.
How does Tilzio detect an agent is done — do I need hooks?
Tilzio listens for the terminal bell (BEL). Claude Code can emit it via its Stop / Notification hooks; any tool that rings the bell works. No proprietary protocol.
Which agents does it work with?
Any CLI agent — Tilzio is agent-agnostic. You run your own Claude Code, Codex, or anything else in a pane; Tilzio does not wrap or proxy them.
Can it show my Claude Code usage limits (5-hour / weekly)?
Yes. The Usage Watcher plugin shows Claude Code and Codex spend plus the 5-hour and weekly limits, right in the app.
Is it a tmux alternative? Do I need a config file to split panes?
GUI drag-and-drop tiling, spaces and grid consoles — no config file, no .tmux.conf. Split by dragging, and your layout and scrollback come back when you reopen.
How is it different from Warp — account or cloud?
No account, no cloud, no telemetry, and the whole app (host + plugins) is open source. Your AI value comes from agents you run yourself — not from a metered backend.
Is it free and open source?
Yes — free, with the host and its plugins fully open source.
Native macOS app or Electron?
Native: Go + Wails v3 with a Svelte front end — one small binary, quick to start and light on memory.
Which macOS version do I need?
Tilzio is a native macOS app and requires macOS 13 or later.
Any telemetry or phone-home?
No. Tilzio makes no network calls of its own; everything stays on your Mac.
Is there a built-in editor, markdown preview, clickable paths?
Yes. A CodeMirror 6 editor with markdown preview lives in panes like a terminal does, and file paths in terminal output are clickable — they open at the exact line.
Can I extend it with plugins?
Yes. Plugins are sandboxed and contribute to fixed UI slots (activity bar, status bar, bottom panel, right column). Install from a zip or URL.
Do I have to use AI agents, or is it a plain terminal too?
It works perfectly well as a plain terminal multiplexer — tiling, spaces, editor, plugins. The AI features only matter when you run agents.