LLM APIs and Providers

Every provider ships a different SDK, but the shape is the same: messages in, choices out, tools callable along the way. The AISL material drills that shape repeatedly — once as a provider tour, once as the tool-calling layer under every agent framework — so that switching providers becomes a configuration change, not a rewrite.

The provider tour

AI Engineering Buildcamp module 1 spends a section on nothing but APIs: the OpenAI Chat Completions API first — messages, roles, streaming — then the same calls against Groq, Anthropic, Amazon Bedrock, Google Gemini, Z.ai, Grok, and locally with Ollama. The point of the tour is not coverage for its own sake; it is seeing which parts of the shape every provider shares.

AI Hero: 7-Day AI Agents Crash-Course day 2 teaches the same shape from the agent side: calling OpenAI, function calling, and the system prompt as the place instructions live — see Prompt Engineering.

Tool calling across providers

Module 3 rebuilds the tour at the layer agents actually depend on: tool calling with the OpenAI Responses API, with Chat Completions, with Anthropic, and with Gemini — all behind one agents library, so the Agents loop runs unchanged on top of any of them. The differences the tour exposed — message formats, tool-schema dialects — get absorbed into the library.

Owning the API itself

The tour ends at hosted providers; Serving Open Models with vLLM on RunPod completes it by serving an open model behind an OpenAI-compatible API — the same shape, self-hosted, billed by the GPU-hour. Deployment details have their own page: Agent Deployment.

Where to go next

• The loop that consumes these APIs: Agents
• Structured responses over any of them: Structured Output