Build an Inference Driver

Coming soon. This tutorial will walk through building a /dev/llm/ inference driver using Pattern 4 (Configured Stream) – the most complex and most powerful pattern in the SDK.

In the meantime, the P4: Configured Stream deep-dive covers the pattern in detail, including how inference maps onto the 6-state machine. The /dev/synth example demonstrates the full P4 lifecycle with a simpler domain.

The first inference driver (Anthropic Claude) is in active development by the core team. Once it ships, this tutorial will walk through building a provider driver from scratch.