Use a custom provider in Codex CLI

What yoink writes

For a provider saved under the profile id openai-api, yoink connect openai-api --to codex --default gpt-5.2-codex produces this, with the key redacted:

model = "gpt-5.2-codex"
model_provider = "openai-api"
 
[model_providers.openai-api]
name = "OpenAI"
base_url = "https://api.openai.com/v1"
wire_api = "responses"
experimental_bearer_token = "<your key>"

name is the provider's display name, base_url is its Responses endpoint, and wire_api is always "responses", which Codex's config reference lists as the only supported value. The key goes in experimental_bearer_token as a literal, because yoink keeps one key in sync across harnesses instead of asking you to export an environment variable. Your other tables and keys in config.toml are left as they were.

Which providers work

Only providers with an openai-responses endpoint can be connected. Among the presets that is openai alone:

In the harness picker those rows show needs an OpenAI Responses endpoint. From the command line, the attempt fails with a message that names the protocol codex needs:

✖ codex: openrouter has no endpoint codex can use (needs one of: openai-responses).

A custom provider gets a Responses endpoint when its /responses route answers yoink's probe, or when you declare it with --protocol openai-responses. See codex is responses-only for the probe details.

Pick a profile id Codex does not reserve

Codex reserves the provider ids openai, ollama and lmstudio for its built-in providers, and its config reference says they cannot be overridden. yoink uses your profile id as the table name, so save the OpenAI preset under a different id before connecting it to codex:

echo "$OPENAI_API_KEY" | yoink add --external --name openai-api --preset openai \
  --models gpt-5.2,gpt-5.2-codex --token-stdin
yoink connect openai-api --to codex --default gpt-5.2-codex

The same applies to a custom provider: avoid naming it ollama or lmstudio if you plan to use it in codex.

Set the default model

--default writes two root keys: model with the model id, and model_provider with the profile id. Codex reads model_provider to choose which provider is active, so a provider you connect without --default sits in the file unused until you pick it. Codex ignores model_provider in a project-local .codex/config.toml, which is why yoink writes the user-level file under CODEX_HOME or ~/.codex.

ChatGPT login versus API key

A model_provider that points at a custom provider overrides your ChatGPT login. When yoink sees a ChatGPT login in auth.json and the provider you just connected is the active one, it prints a notice after connecting, with the full path to your config:

Codex is set to model_provider = "openai-api" in ~/.codex/config.toml, which overrides the ChatGPT login. Run `yoink disconnect openai-api --from codex` or remove model_provider from that file to use ChatGPT.

Switching between several ChatGPT logins is a separate feature that does not touch config.toml; see switch between Codex accounts.

Backups and disconnect

The first time yoink changes an existing config.toml, it saves config.toml.yoink.bak once and never overwrites it. yoink disconnect openai-api --from codex removes the [model_providers.openai-api] table, and removes model and model_provider together when model_provider points at that provider, which hands control back to your ChatGPT login.

FAQ