Which models can I use?

Whatever your model provider resolves. A model provider is any OpenAI-compatible endpoint. The container running your agent sends every model call to it, and the default is OpenRouter.

Your code names a model as a bare id, such as openai/gpt-4o-mini or anthropic/claude-haiku-4.5 on OpenRouter. The container sends that id to its provider as it is, so the id must be one the provider serves. Swapping providers changes no code of yours, but a model id you hardcoded may not exist on the new one. A local Ollama, for example, serves its own tags such as qwen3:4b and refuses openai/gpt-4o-mini. Change the id in your code, or pick another model for the agent in Studio. See Run on Ollama.

The provider is operator configuration. You set it on the Model provider screen, or with MODEL_PROVIDER_BASE_URL and MODEL_PROVIDER_API_KEY on the container. Point the base URL at a local gateway (Ollama, LM Studio, LiteLLM, vLLM) and the key becomes optional.

To read the live catalogue, call stackbone.ai.models.list(). Every entry carries the model id. The provider decides whether it also carries context_length and pricing: OpenRouter reports both, a local gateway usually reports neither.

Read more

• Gateway: the Model provider screen, and every caller in the workspace that resolves through it.
• stackbone.ai: chat, embeddings, images and the model catalogue.
• Choosing a model: where each model id is set and how to compare two of them.
• Getting started: the first run stops until a provider is configured.