An agent's model decides how well it picks tools, follows instructions and handles long tasks, and the provider decides cost, latency, data handling and where it runs. Most teams try more than one: a strong hosted model for hard tasks, a cheaper or local one for simple ones, and a fallback when a provider is down.
These guides start with how to choose a model for tool calling and when to route between small and large models. Then they show how to run the same Promptise agent, with the same MCP tools, on each major provider: Anthropic Claude, Google Gemini, Azure OpenAI, OpenRouter and local models through Ollama. Each covers setup, the provider's quirks with tool calling, and a troubleshooting table, so switching providers is a one-line change.
The code in these guides runs on Promptise Foundry, an open-source Python framework for AI agents and MCP servers.