agentty vs Codex CLI

agentty vs Codex CLI: an honest comparison (2026)

Short answer: Codex CLI is OpenAI's open-source agent with a strong sandbox and first-party GPT support. agentty signs in with the same ChatGPT plan, also runs Claude, Copilot, Kimi and local models, and is ready to type in 47 ms instead of 402 ms from a 23.0 MB binary.

Both are open-source terminal coding agents with an OS-enforced sandbox. Codex CLI is OpenAI's agent, written in Rust and distributed through npm. agentty is a C++26 agent harness shipped as one static binary that works with any provider.

Measured

Both installed the official way on the same machine, fresh home directory, no login. Codex's disk figure includes the Node.js runtime its npm install needs. Median of 10 launches, measured 2026-11-09 against codex-cli 0.167.0.

launch → ready to type lower is better
  1. agentty47 ms
  2. Codex CLI402 ms8.6×
memory while idle lower is better
  1. agentty21 MB
  2. Codex CLI152 MB7.2×
on disk, with runtime lower is better
  1. agentty23 MB
  2. Codex CLI536 MB23×

median of 10 launches · same machine · fresh install, no login · method & all numbers

At a glance

agentty Codex CLI
Built with C++26, one static binary Rust, installed through npm
Ready to type 47 ms 402 ms
Memory while idle 21 MB 152 MB
On disk 23 MB 536 MB
Models Claude, GPT, Copilot, Kimi, DeepSeek, Gemini, Grok, Mistral, Groq, OpenRouter, local Ollama OpenAI models; local through Ollama or LM Studio
Sign in with a plan ChatGPT, Claude Pro/Max, GitHub Copilot, Kimi ChatGPT
Shell sandbox On by default, enforced by the OS On by default, enforced by the OS
Editors Zed and JetBrains over ACP VS Code extension; app server
License MIT Apache-2.0

Where Codex CLI is ahead

  • First-party GPT. Built by OpenAI alongside its models, with Codex Cloud tasks you can pull down and apply locally.
  • Sandbox modes. Clear read-only, workspace-write and full-access modes, with approvals to match.

Where agentty is ahead

  • Not tied to one vendor. Keep your ChatGPT plan and also use Claude, Copilot, Kimi or a local model; switch mid-thread with ^P and ^/.
  • Footprint. 8.6× faster to a usable prompt, 7.2× less memory while idle, 23× smaller on disk, with no npm or Node.js.
  • Retrieval. A local search engine finds the code that matters, so turns stay small.
  • Smart Mode. A flagship model plans; cheaper models do the routine work.

Switching is painless

Your Codex sessions appear in agentty's thread list (Ctrl+J) and you can continue any of them, on GPT or any other model. ~/.codex/AGENTS.md is read as-is, and agentty mcp import --from codex copies the MCP servers from your config.toml. Details in Coming from Codex or Claude Code.

Which should you choose?

  • Codex CLI if you only use OpenAI models and want Codex Cloud.
  • agentty if you want the same kind of sandboxed agent with any model, in a lighter package.

Install in one line

bash
curl -fsSL https://agentty.org/install.sh | sh

Ready to try it? One line, no Node or Python.

curl -fsSL https://agentty.org/install.sh | sh
Get startedGitHub