HagiCode is a proposal-driven AI coding workbench that turns a request into a reviewable proposal, runs it through agent CLIs such as Claude Code, Codex and GitHub Copilot, and keeps the outcome as team knowledge.
Codex is OpenAI's open-source terminal coding agent, signed in with a ChatGPT account or an API key. HagiCode is a desktop workbench that runs Codex as its executor and adds parallel sessions, proposals and AI commits around it.
Codex is OpenAI's coding agent for the terminal: it inspects your repository, edits files and runs commands on your machine. HagiCode is a desktop and web workbench that runs Codex as one of its executors. It does not replace Codex, and it does not sell or host OpenAI models.
If you want a fast terminal agent on your ChatGPT plan, with scripts and CI handled by codex exec, stay with Codex alone. HagiCode is for when you want several Codex tasks side by side, a written proposal before each change, or one workspace across repositories.
OpenAI's terminal coding agent; the same product also has an IDE extension, a desktop app, a web version and cloud tasks
Desktop and web workbench that runs Agent CLIs, Codex among them; not a model or an agent itself
Terminal UI plus codex exec for scripts; installers for macOS, Linux and Windows
Desktop app for Windows, macOS and Linux; Docker Compose container; Microsoft Store edition
CLI repository is Apache-2.0; usage comes with ChatGPT plans (Free, Go $8, Plus $20, Pro from $100 per month, as of October 2026) or API billing
Desktop and Container editions are free; Store edition is free with an optional one-time TurboEngine purchase
OpenAI models through a ChatGPT sign-in or an API key; you choose model and reasoning effort
Hosts no models; uses whatever your Codex install is signed in to; optional OmniRoute model catalog
Subagents inside a session; codex cloud moves work to OpenAI's cloud
Several sessions side by side, each an Agent with a visible status
AGENTS.md via /init, permission levels via /permissions; no proposal archive of its own
OpenSpec proposals: propose, validate, apply, archive; free editions allow 3 concurrent proposals
/review checks uncommitted changes, a commit or a branch; docs suggest Git checkpoints around tasks
AI-composed commits, repository locked while committing, per-repo targeting with Monospecs
One installer, npm or Homebrew command, then sign in with ChatGPT or an API key
Needs Codex installed and signed in first, then HagiCode Desktop on top
Terminal users on a ChatGPT plan, scripts and CI
Parallel tasks, proposal-driven work and multi-repo projects
The Codex repository is published under the Apache-2.0 licence. You can install it with npm install -g @openai/codex, brew install --cask codex, or the standalone installer: curl -fsSL https://chatgpt.com/codex/install.sh | sh on macOS and Linux, and a PowerShell installer on Windows. Run codex in a project and choose Sign in with ChatGPT, or use an API key.
Codex is included in ChatGPT's plans, from Free and Go up to Plus, Pro, Business, Edu and Enterprise, and it shares usage limits with ChatGPT's other work features. An API key is the alternative for automation, billed at standard API rates and kept separate from subscription usage. That makes cost easy to reason about if you already pay for ChatGPT.
/permissions sets when Codex may edit files or run commands without asking, and shows the active sandbox and writable roots. You pick the model and reasoning effort per session. /review reviews uncommitted changes, a commit or a base branch without touching your working tree.
codex exec runs non-interactive commands for CI and repeatable jobs, codex mcp adds MCP servers, --search turns on live web search, --image attaches a screenshot, and codex resume reopens saved chats. Codex can also hand focused work to subagents, and codex cloud moves a task to Codex in the cloud.
codex exec or an API key;/permissions and nothing else;codex cloud.HagiCode sits on top of the CLI. It runs your installed Codex through its SDK-backed integration, so your ChatGPT sign-in or API key, plan limits and billing stay with OpenAI.
Nothing is locked in. Your repositories, AGENTS.md and Codex configuration stay where they are, and you can run codex on the same project at any time.
codex --version.codex once and sign in with ChatGPT or an API key, so it is authenticated before HagiCode uses it.codex command on your PATH.codex in the same folder; nothing needs to be undone.Each Codex session becomes an Agent with its own context and a visible status, so three Codex tasks in three folders are one list instead of three terminals.
A task can start as an OpenSpec proposal that you write, validate, apply and archive. Codex implements within it, and the proposal records why, which codex resume history alone does not.
HagiCode drafts the commit message from what Codex changed and locks the repository while committing. The embedded Code Server editor opens those files beside the conversation.
Monospecs lists the repositories of a multi-repo workspace so a Codex task knows about all of them. Vault keeps reference code and notes available to later sessions, which a new codex run would start without.
Codex's tool calls are shown in the same typed shape as every other CLI, live and in history. OmniRoute can supply a model catalog per Agent CLI, and an optional TokenMonitor readout in the status bar shows today's tokens and cost.
codex directly at any time.Run `npm install -g @openai/codex` or `brew install --cask codex`. The standalone installer for macOS and Linux is `curl -fsSL https://chatgpt.com/codex/install.sh | sh`, and Windows has a PowerShell installer in the repository README at github.com/openai/codex. Then run `codex` and sign in.
The CLI itself is open source under Apache-2.0, but use needs a ChatGPT plan or an API key. As of October 2026 OpenAI lists Codex in ChatGPT Free, Go ($8), Plus ($20) and higher plans; API-key use is billed at API rates.
Yes. Install and sign in to Codex, then choose it as the Agent CLI in HagiCode. Your ChatGPT sign-in or API key and its billing stay with OpenAI.
No. HagiCode runs Codex as its executor and adds sessions, proposals, commits and multi-repo tools around it. You can still run `codex` on the same repositories.
OpenAI provides a PowerShell installer for Windows in the Codex README. Check the Codex documentation for current Windows notes before relying on it.
Pick Codex alone for terminal, solo or CI work on your ChatGPT plan. Add HagiCode when you want parallel sessions, OpenSpec proposals or multi-repo workspaces on top of Codex.
Localized AI code assistant, privacy protection, enhanced efficiency
desktopHero.loading