Skip to main content
Langy runs on the model providers your project has enabled. Every provider except Codex is configured with an API key you paste in model-provider settings; Codex is configured by signing in with your OpenAI account over OAuth. Langy sends all model traffic through the LangWatch AI Gateway, so tracing, rate limits, budgets, and model policy apply to every call.

​
Providers

You add these in model-provider settings, the same drawer the rest of LangWatch uses. Keys are stored encrypted at rest. See Models and bring your own key.

​
Model roles

A role decides which configured model runs for a given kind of work. Langy uses four roles.

​
The model picker

In the Langy composer the picker shows a short shortlist by default, then a More models disclosure with the rest of the models your enabled providers expose.
  • The shortlist is derived, not fixed. It starts with the Langy default, then your current selection, then one representative each of the quick, balanced, and reasoning capability groups. It is not a hardcoded list of names.
  • The More models count is data-driven per project. It is however many other models your enabled providers make available, so it differs between projects.
Named model families you will see when the matching provider is enabled include gpt-5.6-terra (the default), gpt-5.6-sol, gpt-5.6-luna (fast), and claude-opus-4-8.
Codex models appear in the picker only for the Langy and Fast roles. OpenAI’s terms scope the Codex subscription path to coding-assistant harnesses and light AI assists, so Codex is offered for Langy and Fast but not for the playground, evaluations, or workflows. See Codex.

​
Bring your own key

Every provider except Codex is bring-your-own-key: you paste your own API key and Langy runs on your account. There is no LangWatch-provided default key for Langy models. For who pays and how to see the spend, see Token usage and cost.
Also check: Models and bring your own key, Codex, Token usage and cost.

​
Reference