Are you an LLM? Read llms.txt for a summary of the docs, or llms-full.txt for the full context.
Skip to content

Providers

Model credentials, allowed models, and model selections are platform settings. amodal.json has no model field.

Organization Providers

Organization > Model Providers lists the providers in the platform catalog, such as Anthropic, OpenAI, Google, DeepSeek, Groq, Mistral, and xAI. For each provider you can:

  • Enable or disable it.
  • Choose a credential source, such as your own key or Amodal-managed keys when the plan includes them.
  • Add API keys or base URLs where the provider requires them.
  • Select the models available to the organization.
  • Verify the credentials.

A listed model is not proof of working upstream access. The provider can still reject it as unavailable or unauthorized, so verify credentials and run an eval through the deployed agent.

Agent Model

The agent's Models page sets the model, its effort level where supported, the thinking budget, the output-token limit, and the models bound to the simple and advanced tiers. Without a selection, the platform default is google / gemini-3.5-flash.

Hosted runtimes receive the selected model from the platform at deploy resolution. amodal dev accepts --provider and --model overrides.

Tool Schema Compatibility

OpenAI, Anthropic, and Google have separate SDK adapters. Other OpenAI-compatible providers use the OpenAI adapter with a configured base URL. Not every provider accepts every JSON Schema feature.

Native OpenAPI connections send each visible operation's schema as a tool definition. On the Google provider path, the model-facing copy simplifies closed-tuple schemas, because the Google SDK drops their positional item definitions. The runtime validates the actual arguments against the original schema before execution.

Compatibility depends on the provider route as well as the model. Use evals to check the route and schemas your agent uses, including write approval and denial.