Claude Code and Codex desktop apps, compared

An honest table, and the questions worth asking of every tool on it.

Several apps now put a window around Claude Code, Codex and other coding agents. This table lists what each one publishes about itself: what it is, where it runs, which agents it works with and what it costs. Facts about other products were taken from their own sites on 14 November 2026.

AppWhat it isRuns onAgentsPrice
MentatNative desktop app that runs and follows agent sessions, with agent trees, per-agent diffs and a Git clientmacOS now, Windows nextClaude Code, Codex, Grok Build; Antigravity nextFree; Pro is $12 a month, or $96 a year
Claude desktop app, Code tabAnthropic's own desktop app for Claude CodemacOS, Windows, Linux (beta)Claude CodeIncluded with Claude plans
ChatGPT desktop app, CodexOpenAI's own desktop app for CodexmacOS, Windows, LinuxCodexIncluded with ChatGPT plans
ConductorMac app and cloud service for running a team of coding agentsmacOS, cloudClaude Code, Codex, Cursor, OpenCodeSee their site
OrcaOpen-source agent development environmentmacOS, Windows, Linux, iOS, AndroidMany, including Claude Code, Codex and GrokFree, open source
PaseoOpen-source app for running agents from desktop, phone or webDesktop, iOS, Android, webClaude Code, Codex, Copilot, OpenCodeFree, open source
NimbalystOpen-source visual workspace for Claude Code and CodexmacOS, Windows, Linux, iOSClaude Code, CodexFree for individuals
HappyOpen-source client for coding agents, with phone appsmacOS, iOS, AndroidClaude Code, Codex, Grok, GeminiFree, open source

Seven questions to ask

The table shows where the tools overlap. These questions show where they differ. Each has Mentat's answer; put the same question to any other app on the list.

Does it see sessions I start in a plain terminal?
Yes. Mentat reads each tool's own session records, so terminal sessions appear live beside the ones it starts.
When one agent launches another vendor's agent, do I see that?
Yes. A run started with claude -p, codex exec or grok -p nests under the session whose shell command launched it.
Can I see what one agent changed, apart from the others?
Yes. Each diff is rebuilt from that agent's own edits rather than from git diff, so agents sharing a checkout stay apart.
Does it talk to each agent natively?
Yes: stream-json for Claude Code, the app-server for Codex and the Agent Client Protocol for Grok Build. A feature a provider lacks is hidden, not imitated.
Is there a real Git client?
Yes, in Mentat Pro: a commit graph, staging by line, interactive rebase, a conflict resolver, worktrees and undo.
Is it native?
Yes. Written in Rust and drawn on the GPU at 120 frames a second on a 120 Hz display. It is not an Electron app.
Does it need my API keys?
No. Each command-line tool runs under its own login, and Mentat calls no model itself.

For engineers who stay close to the code

Set your agents in motion

Free to use, with no card. Every new account starts with fourteen days of Pro.

macOS 13 or later, Apple silicon. Windows is next. Git 2.45 or later. At least one of Claude Code, Codex or Grok Build, installed and signed in.