docs: Add 4 provider-specific AI Transport getting started guides

[GregHolmes]

Summary

Adds 4 getting started guides for AI Transport, each using a real provider SDK:

• Anthropic - Uses @anthropic-ai/sdk with claude-sonnet-4-5
• OpenAI - Uses openai SDK with gpt-4o
• Vercel AI SDK - Uses ai@^6 with provider-agnostic streaming
• LangGraph - Uses @langchain/langgraph with @langchain/anthropic

Each guide walks through:

• Project setup with provider-specific dependencies
• JWT authentication with Ably
• Defining a send_email tool in the provider's format
• Streaming agent responses via Ably's message-per-response pattern
• Human-in-the-loop tool call approval flow
• A client that displays streamed tokens and handles approval prompts

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 61eb25ef-6293-4752-96d6-578878cd619f

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch AIT-107-Create-AIT-Getting-Started-guides

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[rainbowFi]

I haven't gone through all the details yet, but a couple of quick questions

• what's the motivation for the authenticated users step? We don't do this for the Pub/Sub getting started and it adds further complexity to an already long tutorial
• there are a lot of files here, compared to the existing getting started guides. Is there an easy way that we can package these up for customers to download from this page if they don't fancy doing a lot of copy/paste work?

[GregHolmes]

I haven't gone through all the details yet, but a couple of quick questions

• what's the motivation for the authenticated users step? We don't do this for the Pub/Sub getting started and it adds further complexity to an already long tutorial
• there are a lot of files here, compared to the existing getting started guides. Is there an easy way that we can package these up for customers to download from this page if they don't fancy doing a lot of copy/paste work?

Thank you for the comments @rainbowFi

For the first Q, I went with the auth step because AIT is architecturally different from Pub/Sub getting started guides, in that it always involves a backend agent + frontend client, and the client needs JWTs (not an API key). The human-in-the-loop flow also relies on a clientId from the JWT. I do agree it is heavy, we could slim it down by linking to a shared auth snippet or providing a ready-to-run auth server, so the guide focuses on the AI-specific parts, but this would complicate the maintenance of these guides for anyone in the future. What do you think?

For the second Q, I agree, 3-5 files per guide is a lot of copy/paste. The simplest option could be where ever we store the feature guides code, we store the getting started guides in there too. It's currently a PR in docs, but I think it's still up for debate on where it's going. Something @mschristensen and @m-hulbert were discussing. We already have runnable examples in /examples/ for the message patterns, but these don't use actual providers. So unless we ask the user in the flow of the example to provide their API Key for the specific provider, the examples wouldn't otherwise be feasible.

[rainbowFi]

Thanks @GregHolmes

For the first question, I think it's better to have everything on one page rather than linking to some common example page. We don't want people to have to click through multiple pages to get this single "Getting Started" working, and I agree with you that the auth is necessary.

For the second one, let's say the code lives in whatever location we pick for the feature guide code, because it's the same problem of people needing a quick way to access it. I'm not going to gate approval of this PR on that discussion coming to a conclusion. Can you make sure that the work is tracked somewhere?

In terms of the actual content, I've put one request inline and tested the Anthropic example. I haven't reviewed the code for quality, if you have any concerns then I suggest you ask an engineer to take a look.

[rainbowFi]

Is this strictly required for anything else that is covered in the getting started guide? The way we're going to track AI transport usage will add an agent/human identifier to the connection parameters, so I'd rather not start people doing this through the JWT.

[rainbowFi]

Is this claim strictly required for anything in this Getting Started Guide? The plan for AI transport usage tracking is to specify human or agent as part of the connection parameters, so I don't want people also doing that through the JWT.