01
Model catalog
When third-party model names differ from official ones, a model_catalog_json is injected so the client can list them.
Client support · OpenAI Codex
ChatGPT（Codex）
Codex keeps its model catalog, reasoning depth and context length in config.toml. AiSwitch writes those fields for you, so switching endpoints no longer means reading docs and guessing.
What you get after takeover
02
Reasoning depth
Depth values are normalized before writing, so the client never hits an unknown enum and crashes.
03
Stale locks
Leftover thread write locks are cleaned after a switch or a force quit, so archiving keeps working.
Three steps
Pick ChatGPT
It is the first row in the sidebar.
Add endpoint and key
Paste the base URL and key, then pull the model list.
Click "Use it"
Takes effect after the client restarts; revert in one click.
FAQ
Fewer models after switching?
The catalog was not written. Hit "Refresh models" and switch again.
Does it rewrite my config.toml?
Yes, but it is backed up first and restored when you switch back to official.
Point this client at your own API today
One installer for Windows and macOS, 7.5 MB, no account needed.