Codex

Connect the Codex CLI (OpenAI’s codex) to Routero by defining a custom model provider in its config. Codex then sends every request through your gateway.

Codex supports two wire formats — chat (/chat/completions) and responses (/responses). Routero supports both; responses is recommended for Codex.

What you must set

In ~/.codex/config.toml:

model_provider = "routero"
model = "openai/gpt-5.5"

[model_providers.routero]
name = "routero"
base_url = "https://api.routero.dcsmartvision.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "responses"

Then provide your Routero key as the API key:

export OPENAI_API_KEY="YOUR_ROUTERO_KEY"
Setting Value / meaning
model_provider the name of the [model_providers.*] block below (routero)
model any model Routero serves (e.g. openai/gpt-5.5)
env_key OPENAI_API_KEY — the env var Codex reads your Routero virtual key from and sends as Authorization: Bearer
base_url https://api.routero.dcsmartvision.com/v1
wire_api responses (recommended) or chat
OPENAI_API_KEY your Routero virtual key, not an OpenAI key

For a custom provider, Codex sends the key named by env_key. Without env_key, the key is never attached and Routero returns 401 Unauthorized: No api key passed in.

preferred_auth_method = "apikey", sometimes cited for API-key auth, is not recognized by recent Codex versions (verified against Codex CLI v0.157.1, which warns it is an unknown config key and ignores it). On older releases it may still be accepted; it is harmless to remove.

Create a key for each developer

In the dashboard, open API Keys and create a virtual key per developer — scoped to their team, limited to the approved models, with an optional budget.

→ Calling the API for the base URL and auth model. → Claude Code and Cursor for the other agents.