Originally created by: sosidudku1
Current behavior
When a user configures an OpenAI-compatible cloud provider (custom base URL, e.g. vLLM or LM Studio), the setup wizard fetches {baseUrl}/v1/models and shows a model list to pick from (#41). After the wizard completes, that list is gone for good:
openai-compatible provider: the one already selected (chatModelsForProvider in src/tui/llm-panel/llm-panel-row-builders.ts). Pressing Enter on it re-selects the same model./model only jumps to the LLM tab; it takes no argument and opens no picker (src/tui/commands/slash-command-handler.ts).https://api.openai.com, clobbering the custom endpoint.Effectively the user is locked into the model chosen during first-time setup unless they re-run the whole wizard or edit config.json by hand.
Expected behavior
A model picker that can be reopened at any time, which is the common pattern in agent CLIs: /model and Enter on the current model row should open the provider's model list, with the current model marked, and switching should take effect mid-session without re-entering the API key or the base URL.
Implementation note
The building blocks already exist: fetchOpenAiCompatModels / its 1h cache (added in [#41]) and the wizard's pick UI. The main work is exposing that picker outside the wizard: trigger it from the cloudChatModel row and from /model, and route the selection through the existing onProvidersSelectChatModel callback, which already persists and applies the change.
Related
While here, the configure-mode wizard should prefill the stored base URL instead of defaulting to https://api.openai.com on an empty field; today a quick re-run can silently break a custom endpoint.
Ticket changed by: Ooooze