Codex
Codex is OpenAI's CLI coding agent. It speaks the OpenAI wire format and reads its provider settings from a TOML config file.
Support state: auto-configurable. routerly clients configure writes
this file for you.
Config file: ~/.codex/config.toml (TOML)
Wire format: OpenAI
Auto-configure (recommended)
With the CLI installed and logged in:
routerly clients configure codex --project <your-project>
This mints a project token (or reuses one you pass with --token), backs
up your existing ~/.codex/config.toml, and adds the block below. Any
other provider or setting already in the file is left untouched.
Minting a new token requires project:write permission on the target
project (use --token <existing-token> to skip this).
Manual configuration
Open ~/.codex/config.toml and add:
model_provider = "routerly"
[model_providers.routerly]
name = "Routerly"
base_url = "http://localhost:3000/v1"
wire_api = "responses"
experimental_bearer_token = "sk-rt-YOUR_PROJECT_TOKEN"
base_url includes the /v1 suffix. Replace http://localhost:3000/v1
with your Routerly instance's URL and sk-rt-YOUR_PROJECT_TOKEN with a
project token (create one on the Projects
page, or let clients configure mint one for you).
Undo
Every configure run prints a Backup ID. Restore the file to its exact
prior state with:
routerly clients undo <backupId>
See CLI: routerly clients for the full command
reference, including inspect (check current status) and doctor (check
service reachability).