Native, not a web page in a box. Codz is AppKit end to end. The transcript recycles its rows, so a five-hour turn with hundreds of tool calls scrolls like a short one.
Built for the Mac
NewAn issue tracker in every project
One command center for every coding agent.
Claude Code, Codex, Cursor and OpenCode side by side in a native Mac app, with every diff, file, approval and dollar in view.
This is the Codz app, running in your browser with a scripted agent. Ask for a change:
Works with the agents you already use
- Claude CodeAnthropic
- CodexOpenAI
- CursorCursor Agent
- OpenCodeOpen source
- OpenRouterAny model
- DeepSeekV4 Pro
01Agents
Every agent, one conversation.
Run Claude Code, Codex, Cursor and OpenCode side by side, and hand a thread from one to another without losing the context.
How the model picker works
02Review
Review every change before it lands.
Diffs, files, Git and approvals sit beside the chat, so you decide what lands and what runs.
See what’s new in the rail
03At a glance
In your menu bar, on your desktop, in your terminal.
See which agents are working, how close you are to a limit and, with Pro, what they cost, without opening a window.
codz agents --json
04Remote
Your Mac, in your pocket.
Codz Remote steers your Mac’s agents from iPhone and iPad, so an approval never waits for you to get back to your desk.
Included with Pro
The calm way to run coding agents.
Your logins stay yours. Codz drives the Claude, Codex, Cursor and OpenCode command-line tools you already signed in to, with each one’s own account and limits.
Bring your own agents
Nothing invented. Token counts and costs come from each provider’s own logs. Where a provider reports context rather than billing, Codz says so instead of guessing.
Honest usage
Private by default. The free app works entirely on your Mac. Sync and Remote are opt-in, and Remote’s relay only ever carries ciphertext.
Local first
Every change reviewable. Edits land as diffs you can open, commands wait when your trust level says they should, and questions come back as cards you answer in place.
You decide
Many agents at once. Queue follow-ups, run tasks in their own worktrees, and let the sidebar’s bell tell you which chat is waiting on you.
Parallel by design
Stay in control
Use the best model for every task
Put Claude, GPT, Cursor, DeepSeek and OpenRouter models in one loadout and switch with a key.
About the loadout
Know what every agent costs
Tokens, cost estimates and plan windows per provider, per model and per day, from local logs.
About usage
Ship from the chat
Commit, push and open the pull request from the Git panel once the work is done.
Get Codz
