Provider Assets

API and local profiles share the LlmProviderDefinitionBase configuration model while supplying endpoint, authentication, and response parsing appropriate to their provider family. Codex and Claude Code are editor-only subscription connections, not runtime API profiles.

Built-in providers

ProviderEndpoint styleNotes
OpenAI/v1/chat/completionsReference shape for OpenAI-compatible providers.
OpenAI CompatibleConfigurable OpenAI-style base URL and chat endpointUse when a service implements the compatible chat-completions request/response shape.
DeepSeekOpenAI-compatible chat completionsSame general transport shape with a different base endpoint.
GeminigenerateContent styleUses Google-style API key header rules.
Ollama/api/chatLocal or remote Ollama endpoint, no remote cloud key by default.
Custom JSONUser-defined method, body, and response pathConfigure this for a custom JSON API; choose OpenAI Compatible when the endpoint follows that common schema.
Interactive preview: Profiles, recreated from the product source with fictitious example values. Select profiles to inspect their stored fields; no provider is contacted. Open the full preview.

Model selection and duplicate profiles

Fetched model IDs appear in a searchable, alphabetized list. Search by any part of the full ID; slash-containing IDs remain intact and long IDs expose a full tooltip. Choosing a result applies it to the profile; filtering alone does not change the model. Custom model names can also be entered manually. Editing a profile preserves its asset GUID and credential scope. Duplicating creates a new asset identity and does not copy credentials or connection-test state.

What provider assets store

  • Display name and description.
  • Base URL and endpoint rules.
  • Default model name.
  • Transport and reliability settings used by request execution.
  • Request header behavior.
  • Provider GUID used for local key lookup.

What provider assets do not store

Local API keys are not serialized into the shared asset file. That boundary is intentional and matters both for version control and for honest public documentation.

Editor-only subscription connections

Codex and Claude Code profiles use native tools that the editor user installs and signs into. They are not part of player builds and do not store subscription tokens as API keys. Use an HTTP/API provider for runtime requests.

Built-in provider surface

AI Providers 2.2 setup checklist before configuring a provider

The built-in provider tabs show the supported provider families and the editable asset fields each provider exposes.