Models, Providers, and Profiles
A provider is a service that runs models, and a model is the specific one the agent thinks with. Wayland separates the two: you connect a provider once with credentials, then choose among its models. Different assistants and conversations can use different models, so you can pair a strong reasoning model where it matters with a faster one elsewhere.
The catalog and the engine-native five
Section titled “The catalog and the engine-native five”The desktop catalog lists 32 providers, from Anthropic and OpenAI to OpenRouter, Groq, Mistral, and local and regional services. Five of these are engine-native, meaning the engine speaks to their APIs directly: Anthropic, OpenAI and its compatibles, Google Gemini, AWS Bedrock, and Google Vertex AI. The OpenAI-compatible path is wide: it covers DeepSeek, Ollama, and other services that present an OpenAI-shaped API. The broader desktop catalog builds on these five plus additional integrations.
ProviderCompat
Section titled “ProviderCompat”Each provider has quirks: a different field name for the token limit, different rules about merging messages, schema sanitization for Bedrock. The engine handles these through a ProviderCompat layer, which is configuration rather than hardcoded branches. Each provider type has sensible defaults, and you can override any field. Adding or adjusting a provider is a config change, not a code change, which is what keeps the catalog wide.
Credentials
Section titled “Credentials”You add a cloud provider by entering its credentials in the Models settings, where the desktop also offers one-click and JSON import for getting keys in quickly. The engine reads credentials from its config file and from environment variables, with flags able to override per run. See Add a Cloud Model Provider and Use a Local Model.
Profiles
Section titled “Profiles”A profile is a named bundle of provider and model settings you can switch to as a unit. Profiles support an extends field, so one profile can inherit from another and change only what differs. A profile for a fast local model and a profile for a frontier cloud model can each be a one-flag switch. On the command line, --profile selects one; the desktop offers the same model selection per conversation.