Codex

Install the client

Codex 0.166.0 was exercised with two requests, streamed output and a client-side tool/result round trip against a synthetic local response. This is client/protocol evidence, not live provider or production billing verification. Install this version from the official npm package; review changes before updating.

npm install -g @openai/codex@0.166.0
codex --version

Before you begin

Open a fresh terminal, configure a scoped key and list the models available to your account before using the client. The shared setup includes Bash and PowerShell commands.

• Credentials and model check

Use a separate profile

Save the generated configuration as ~/.codex/notholy.config.toml; on Windows use %USERPROFILE%/.codex/notholy.config.toml. Set NOTHOLY_API_KEY privately, then launch codex -p notholy. This separate file preserves your default provider. Disable multi-agent namespaces and hosted web search for this compatible Responses path.

• Codex configuration reference

Responses compatibility

Codex requires Responses, not Chat Completions. NotHoly provides stateless Responses with caller-executed function tools. Hosted web search, remote compaction and stored response retrieval are not interchangeable with this interface. A configuration example alone is not evidence that every Codex feature works.

Client configuration

Select a published model to generate configuration. Confirm your account can use it with GET /v1/models before sending a request.

model_provider = "notholy"
model = "notholy-pulse"
web_search = "disabled"
model_reasoning_summary = "none"

[features]
multi_agent = false

[model_providers.notholy]
name = "NotHoly"
base_url = "https://api.notholy.com/v1"
env_key = "NOTHOLY_API_KEY"
wire_api = "responses"
request_max_retries = 0
stream_max_retries = 0
supports_websockets = false

Configuration reference: a generated snippet is not a full client certification. Check the compatibility notes above. Keep API keys in your private environment.