Use Continuum’s hosted inference with the ChatGPT desktop app’s Codex

What the ChatGPT desktop app actually runs

Codex inside the ChatGPT desktop app is not a reimplementation and not a hosted service with a local skin. The app ships the genuine CLI binary and drives it in app-server mode, so the thing answering your prompts in that window is the same program you would run from a terminal.

That is a genuinely good starting position, and it is why this guide exists rather than saying "you cannot". It is also not the end of the story, because reading your configuration and letting you select what is in it are two different things.

Why the in-app route is a trap

Adding a custom provider and then picking its model from the app’s own model picker is the obvious move, and it does not work. The failure is not a permissions or credentials problem; it is that the picker does not carry provider identity, so the app changes which model name it asks for without changing who it asks.

The reproduced symptom is worth knowing by sight, because it sends people to check their API key when the key is fine. On Codex Desktop 26.730.8199, selecting a custom model produces a "Model changed…" confirmation, immediately followed by a failure reading The "<model>" model is not supported when using Codex with a ChatGPT account. The app switched the label and kept the provider, so it asked the ChatGPT backend for a model that backend has never heard of.

The setup that works: a Codex profile

Profiles are Codex’s own mechanism for a named overlay on top of the default configuration. That is exactly the shape of this problem: you want Continuum available in the config home the desktop reads, without it being what the desktop picks up by default. Nothing here touches the app’s ChatGPT login.

One prerequisite before the steps: this needs Continuum Plus or higher. Inference keys do not mint on the Free tier, so the Create key button stays disabled until you subscribe, and the profile below is inert without a key.

[model_providers.continuum]
name = "Continuum"
base_url = "https://continuumcode.ai/v1"
env_key = "CONTINUUM_API_KEY"
wire_api = "responses"

[profiles.continuum]
model_provider = "continuum"
model = "gpt-5.6-sol"

export CONTINUUM_API_KEY="cont_sk_..."
codex --profile continuum

What this gets you, and what it does not

It is worth naming the result precisely rather than letting the guide’s title over-promise. As of 13 August 2026, "Continuum in the ChatGPT desktop app" resolves to Continuum in the bundled Codex CLI, launched from a terminal, while the desktop GUI stays on your ChatGPT account. The config home is shared, the binary is the same one; the window is not the surface that reaches Continuum.

Three things not to do

• Do not use the requires_openai_auth = true trick. Making a third-party provider claim to be OpenAI-authenticated is a fake-auth hack, not a configuration, and it earns you a setup that breaks in ways nobody will help you debug.
• Do not use model_catalog_json. It is a full replacement rather than a merge, so adding one custom model deletes the entire built-in catalog for that process.
• Do not use a separate CODEX_HOME. It looks like clean isolation and costs you your chat history, because session state is tied to the active home.

Each of those is attractive precisely because it looks like it routes around the picker problem. None of them do; they trade a visible limitation for an invisible one.

When it does not work

Start every diagnosis in a terminal. The desktop app’s environment is the hardest part of this setup to reason about, so proving the profile works from a shell first removes the largest variable in one step.

On wire_api, which is not negotiable

The configuration reference gives "responses" as the only supported value, and the binary enforces it rather than warning. The error is unusually helpful, so it is worth reproducing exactly:

Error loading config.toml: `wire_api = "chat"` is no longer supported.
How to fix: set `wire_api = "responses"` in your provider config.
More info: https://github.com/openai/codex/discussions/7782
in `model_providers.continuum.wire_api`

curl -s https://continuumcode.ai/v1/models \
  -H "Authorization: Bearer cont_sk_..."

curl -s https://continuumcode.ai/v1/responses \
  -H "Authorization: Bearer cont_sk_..." \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.6-sol","input":"say hi"}'

The paid lane serves claude-opus-5, claude-fable-5, claude-sonnet-5, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, grok-4.6, gemini-3.5-flash, deepseek-v4-flash, glm-5.3, and muse-spark-1.2, read from the live gateway on 17 August 2026, alongside a free lane. Personal API keys are a paid-plan feature: Plus at $25/mo, Max 100 at $100/mo, Max 200 at $200/mo, or Ultra at $500/mo. The full table, free lane included, is in the Codex CLI guide.

Questions people ask

Keep reading