Skip to content

README.md

Latest commit

 

History

History
65 lines (48 loc) · 1.73 KB

README.md

File metadata and controls

65 lines (48 loc) · 1.73 KB
Raw

API Key Service Tier

This opt-in feature exposes the desktop Fast/service-tier selector when Codex is using API-key authentication with an OpenAI-compatible provider.

It is intended for providers that wrap the OpenAI Responses API and understand Codex's serviceTier request setting. It does not grant OpenAI Fast mode credits and does not bypass ChatGPT-account entitlement checks for the official OpenAI service.

Enable

Add the feature id to linux-features/features.json:

{
  "enabled": [
    "api-key-service-tier"
  ]
}

Then rebuild the app:

./install.sh

Keep the model name as the upstream model name, for example:

model = "gpt-5"
model_provider = "openai-compatible"
service_tier = "fast"

[model_providers.openai-compatible]
name = "OpenAI-compatible"
base_url = "https://provider.example/v1"
wire_api = "responses"

Provider-specific API keys should be configured according to the provider's normal Codex setup. Do not encode the tier into the model name unless the provider explicitly documents such a model alias.

Behavior

  • API-key-authenticated hosts are allowed to show service-tier controls.
  • If an API-key host's active model has no serviceTiers metadata, the UI synthesizes one fast option so the selector can send serviceTier: "fast".
  • ChatGPT-authenticated hosts still use upstream account requirements for official Fast mode and do not receive synthetic service-tier metadata.

Risks

The provider must accept and implement the serviceTier request setting. A provider that rejects unknown fields may return an API error; a provider that ignores unknown fields may show the UI without changing latency.

Test

node --test linux-features/api-key-service-tier/test.js