Code Mode, the coding agent built into ibl.ai's Agentic OS, now runs on OpenAI's Codex or Anthropic's Claude Code as well as on ibl.ai's own models — one row in the Code panel switches between them, and on ibl.ai you own all the code and the data, including the code the agent writes.
Picking an agent is a single control. Everything around it stays the same: the same approval cards, the same workspace folder on your computer, the same skills.
Switching Code Mode between ibl.ai, Codex and Claude Code, and building a site on each.
For what Code Mode does on its own, see the original Code Mode announcement; for driving it from a phone, see Code Mode From Your Phone.
Which coding agents can Code Mode use?
Three, chosen on the Agent row in the Code panel, and the choice is the whole switch.
| Agent | What it needs | What it runs on |
|---|---|---|
| ibl.ai | No sign-in | Your organization's models and credits |
| Codex | A ChatGPT plan, or an OpenAI API key | Your own subscription |
| Claude Code | A Claude plan, or an Anthropic Console account | Your own subscription |
The line under the row always says what is still missing — Not installed, Not signed in, or the account you are signed in with. Check Again re-reads it.
How do you sign in to Codex or Claude Code?
You don't, separately. Both reuse a sign-in you already made somewhere else, which is the point — Code Mode never asks for your OpenAI or Anthropic credentials.
Codex on macOS lives inside the ChatGPT desktop app. The download at openai.com/codex is ChatGPT.dmg, and there is no separate codex command to install.
Sign in to Codex inside the ChatGPT app with the account that carries your plan, press Check Again, and the line shows how you are signed in. A Codex CLI login counts too.
Claude Code signs in from a terminal, because its official install is the command-line tool. Run claude once, pick your Claude account, finish in the browser, and Check Again shows Claude Code ready.
One thing worth knowing before you go looking: the Claude desktop app's own sign-in does not carry over. It has to be the CLI login.
Does the model picker change with the agent?
Yes — it follows whichever agent is selected, and the choice is saved on that computer.
- On ibl.ai it lists your platform's own providers.
- On Codex it lists what your ChatGPT plan offers, the GPT-6 and GPT-5.6 families.
- On Claude Code it lists the Claude models: Opus 5.5, Fable 5.1, Sonnet 5 and Haiku 4.5.
The same conversation can continue on a different agent in the same workspace. Start a build on ibl.ai, hand the next turn to Claude Code, and the work carries on in the same folder.
Do approvals still apply to a third-party agent?
They do, and they map onto each agent's own permission model rather than sitting beside it.
Ask Me raises a card before every file edit and every command. Automatic lets the agent work without asking, inside the workspace folder you chose.
On Codex, Ask Me maps onto Codex's own read-only mode with approvals, and Automatic onto full access. The Workspace section shows the folder, with Open in Finder, Select Workspace and New Workspace; Phone Access pairs a phone by QR code to drive Code from another room.
What is actually installed on your machine?
Its own copies, deliberately. The desktop app installs its own Node runtime and the Codex and Claude Code adapters in the background on first launch, and runs them by absolute path — nothing on your PATH is used.
The agents reuse the logins you made elsewhere, run inside a sandbox bound to the workspace folder, and receive the same ibl.ai guidance and skills as the built-in agent.
So a third-party coding agent does not arrive as an unmanaged process on your machine. It arrives on the same footing as ours: same sandbox, same workspace, same approvals, same skills.
What does a Code session look like?
Like any other chat. Ask for a website and the agent offers the default template, asks what the site should do and who it is for, then builds it with every tool call visible in the same bubble — and says so when a task will take a few minutes.
Full reference, including the adapters and the sign-in states: the /iblai-vibe-os-agent-code-mode skill. The desktop app ships for macOS and Windows from iblai/os, whose source is MIT-licensed.
Why does this matter beyond convenience?
Because it is what model-agnostic means when you follow it all the way down.
Most platforms let you pick which LLM answers a chat turn. Agentic OS lets you pick whose coding agent does the work, on whose subscription, against a workspace on your own computer — and switch mid-conversation.
That is the same architecture as the rest of the platform: ibl.ai is LLM-agnostic end to end, bring any model and switch anytime, with no per-seat pricing and deployment in your cloud, on-premise, in GovCloud or fully air-gapped.
1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.
Want your team coding with agents on your own machines?
We can show Code Mode running end to end — ibl.ai, Codex and Claude Code, with approvals and a workspace you keep. Book a 30-minute demo or talk to the ibl.ai team — ibl.ai is family-owned and operated from New York, NY.