OrcaRouter provider support for AI Engineering from Scratch

AI Engineering from Scratch is an open, MIT-licensed reference manual and curriculum: 523 lessons across 20 phases, built on the observation that most AI material arrives in scattered pieces, leaving learners able to ship a chatbot without explaining its loss curve. Its answer is that you build AI end-to-end by hand, and every lesson ships a reusable artifact: a prompt, a skill, an agent, or an MCP server.

That build-by-hand workflow is why the APIs & Keys lesson reduces every provider call to one pattern (endpoint, API key, request, response) while noting that Anthropic, OpenAI, and Google each keep their own SDK and schema. So learners who compare models must adopt a new client each time. An optional provider speaking one common API would give them more model choice without more SDK churn.

Proposal: OrcaRouter as an optional provider

I'm an engineer on the OrcaRouter team. I'd like to propose documenting OrcaRouter as an optional provider that learners and skill users can point their local builds at, alongside (never replacing) the Anthropic, OpenAI, and Google examples already in the curriculum.

OrcaRouter is a model gateway that exposes an OpenAI-compatible API and uses standard API-key authentication, so it fits the exact "endpoint + key + request" shape the APIs & Keys lesson teaches. A natural integration point would be an optional section or skill output extending that lesson, where a learner sets an ORCAROUTER_API_KEY and base URL and calls one endpoint to reach models across providers. Nothing is implemented or tested yet; this is a proposal for maintainers to weigh, not a claim of finished code.

Capabilities most relevant here:

• Multiple chat, reasoning, image, and video models behind one endpoint: the lesson's "one pattern" framing, directly.
• Automatic model routing and provider failover: a live, configurable echo of the routing and orchestrator-worker patterns Phase 14 teaches.
• Usage tracking and budgets: useful to learners who watch costs climb while running agent loops.

OrcaRouter is already part of the wider open-source ecosystem; it is used or referenced by projects such as Dify, goose, promptfoo, and OpenCode / models.dev, so documenting it here would also help learners who meet it again inside those tools.

Transparency note: OrcaRouter runs an optional open-source partner program in which approved OSS projects can receive a 5% revenue share from OrcaRouter usage attributed to their integration. Participation is optional, is not a condition for adding the provider, and I would gladly follow this repo's disclosure and governance expectations either way. More context: https://www.orcarouter.ai/built-with.

I'd welcome the maintainers' thoughts on whether an optional OrcaRouter example fits the curriculum and where it belongs. With your approval I would be happy to open a follow-up PR.