Codex API Setup

Codex is OpenAI's terminal coding assistant. It connects to UnioAPI through a custom provider via POST /v1/responses.

1. Install Codex CLI

Skip to section 2 if it is already installed. Requires Node.js (18 or newer):

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

2. Add the UnioAPI provider

Edit the config file — ~/.codex/config.toml on macOS / Linux, %USERPROFILE%\.codex\config.toml on Windows (create it if missing):

model = "gpt-5.6-sol"
model_provider = "unioapi"

[model_providers.unioapi]
name = "UnioAPI"
base_url = "https://api.unioapi.com/v1"
env_key = "UNIO_API_KEY"
wire_api = "responses"

Three key points:

wire_api = "responses": since February 2026 Codex only supports the Responses transport (wire_api = "chat" was removed; older guides are outdated). UnioAPI serves POST /v1/responses natively.
Top-level model_provider must match the [model_providers.<id>] id; openai, ollama, and lmstudio are reserved ids.
env_key names the environment variable that holds your key; the key never lives in the file.

3. Export the key and launch

macOS / Linux

SHELL_RC="$HOME/.zshrc"; case "$SHELL" in *bash*) SHELL_RC="$HOME/.bashrc";; esac
echo 'export UNIO_API_KEY="your UnioAPI key"' >> "$SHELL_RC"
source "$SHELL_RC"
codex

Windows (PowerShell)

setx UNIO_API_KEY "your UnioAPI key"

Reopen the terminal, then run codex. A normal reply means you are connected. To switch models, change the top-level model to any OpenAI-protocol model from GET /v1/models.

Troubleshooting

Symptom	Check
Provider config error on launch	`wire_api` must be `responses`; top-level `model_provider` must match the block id.
401 authentication failed	Confirm `UNIO_API_KEY` is exported and matches `env_key`. On Windows, reopen the terminal after `setx`.
400 parameter rejected	Capabilities like `background` are out of scope — see POST /v1/responses.
404 model not found	Top-level `model` is not an available model identifier.