Core concepts

Providers

The runtime is provider-independent. Provider configuration and the curated model catalog are exposed through the API; runner and execution code should depend on capabilities rather than vendor-specific logic.

Current providers

• Anthropic;
• NVIDIA;
• Ollama;
• OpenAI;
• Together.

The curated model catalog currently contains deployments for Anthropic, OpenAI, and Together. NVIDIA and Ollama are valid configurable providers, but they do not yet expose selectable curated deployments. Their provider API response returns a zero model count and an explicit catalog note until model discovery or supported deployments are added.

Future providers

Future providers may include Gemini and enterprise models. Adding a provider should not require provider-specific behavior in the runner.