The payoff of the last two labs. The package is installed, the token's in your environment — now call the real SDK: a single chat, models running in parallel, and a full multi-agent flow with a human in the loop. This is genuinely what testing FireSDK looks like.
Quick Refresher
Same REPL, real SDK calls now.
Same >>> / ... Python REPL from the earlier labs. Every call here goes through fire_sdk — the real package you installed — and every response you'll see is a genuine capture from the live Fire API, replayed here exactly. (The placeholder token stands in for a real one; when you do this against the live API, you'll mint your own first.)
A couple of exercises here pause mid-flow waiting on you (run.wait(), awaiting_human) — that's not a mistake, it's the actual multi-agent flow doing its job.
Done
You just tested every layer FireSDK has.
L1 (a single chat call), Species (typo-proof model names instead of raw strings), L2's cousin at the parallel layer, and v2 — a real multi-agent flow that paused for your input mid-run and resumed with it. That "run" object staying alive across a pause is the one piece of this whole track that doesn't reduce to a single request/response round trip. Everything else you've learned — curl, Python's urllib, JS's fetch — was building toward being able to recognize exactly what FireSDK is doing under the hood when you call it.