Codex CLI with ShareQuota
Codex speaks the OpenAI Responses dialect, and the proxy serves it on its own path. One provider block in config.toml and one key in the environment.
Codex CLI is configured through ~/.codex/config.toml, where a model_providers table names a base URL, the environment variable that holds the key, and the wire protocol. Point it at the local endpoint with wire_api set to responses, export the sq- key under the name you chose, and Codex runs against the ChatGPT plan you signed in to — or any other provider, by changing the model id.
Setup
- 1Start the proxy and connect a planRun the binary, open the dashboard, and sign in to your ChatGPT account, or paste an OpenAI API key.
- 2Add the provider blockPut the model_providers section into ~/.codex/config.toml and set model_provider to it. The dashboard’s one-click switch writes exactly this section.
- 3Export the key and run codexExport your sq- key under the env_key name from the block, then start codex as usual.
What to set
model_provider = "sharequota"model = "chatgpt/gpt-5.4"[model_providers.sharequota]name = "ShareQuota"base_url = "http://localhost:20130/v1"env_key = "SHAREQUOTA_API_KEY"wire_api = "responses"Worth knowing
- wire_api must be responses: Codex speaks the Responses dialect, and the proxy serves it at /v1/responses rather than translating chat completions.
- The model id decides the provider — chatgpt/ lands on the ChatGPT connection, claude/ on Claude, and so on.
To undo it
Remove the entries above, or turn the integration off in the dashboard and the backup is restored. Nothing else was written.
Questions
Yes. The proxy signs in to the ChatGPT account you connected with OAuth, and Codex spends that plan through the local endpoint.
Yes — set the model to claude/claude-sonnet-5 and the proxy translates the Responses dialect to the Anthropic one, stream included.
Checked against the tool’s own documentation on 2026-09-13. Settings move; if a field has been renamed, the docs link above is the source.