AI model provider tools give agents programmatic access to language models, embedding models, image generators, and speech services. These range from official provider SDKs (OpenAI, Anthropic, Google) to local inference tools (Ollama, llama.cpp, LM Studio) that run models on your own hardware.

In multi-agent architectures, model provider tools often serve as the "brain" that other specialized agents call. An orchestrator might use a fast, cheap model for routing decisions while delegating complex reasoning tasks to a more capable model — all through a unified model provider interface.

Local inference tools deserve special attention: Ollama in particular has become the standard for running open models (Llama, Mistral, Phi, Gemma) locally. This matters for privacy-sensitive workloads, offline operation, and cost control at scale. The tradeoff is hardware requirements and latency vs. API-based models.