CT Feature (WIP)

CT is a provider switcher plus local proxy layer that lets Claude-compatible workflows use different model backends with one command.

What CT does

• Defines provider aliases (named providers) you can switch to with ct <provider>.
• Supports agent/provider mappings so teams can route agent types to specific providers.
• Includes a custom local Codex proxy that exposes an Anthropic-compatible endpoint while using your ChatGPT/Codex subscription behind the scenes.
• Lets you use your Codex subscription through Claude workflows by switching to the codex provider.

Provider menu (TUI)

This is the provider selection view where generic providers and codex are managed/switched.

Core usage

# Add a named provider alias
ct provider add openrouter generic https://openrouter.ai/api/v1 $OPENROUTER_API_KEY

# Switch by alias
eval "$(ct openrouter)"

# Switch to Codex via local proxy
eval "$(ct codex)"

# Full Codex setup (enable provider + daemon + auth)
ct codex --onboard

# View status
ct status
ct codex --status

Claude using CT

This shows Claude actually invoking the tool and running CT-managed shell commands.

Codex proxy behavior

When codex is active, CT manages env/config so Claude-compatible clients point to the local proxy endpoint (http://127.0.0.1:4096/v1) and use managed credentials/auth flow.

Local dev note

In this repo, run with the local token file:

CODEX_PROXY_TOKEN_FILE=/Users/chrismck/Code/codex_in_claude_code/.tokens.json npm run start

Acknowledgements

• Some of this implementation is based on numman-ali/opencode-openai-codex-auth: https://github.com/numman-ali/opencode-openai-codex-auth
• We also borrowed from opencode's PKCE OpenAI Codex login approach: https://github.com/anomalyco/opencode
• Thank you to z.ai GLM for coding most of this, and Codex for fixing what GLM coded.

Status

This feature is a work in progress. Current effort is focused on improving tool-call usage and reliability in the Codex proxy specifically.