nx_companion_compose_test.nx

nx_companion_compose_test.nx -- end-to-end integration smoke proving the conductor + parallel-companion substrate composes with the existing Nishi AI LLM stack. Wires together (all bits-up NishiLang -- no external imports): Other-agent stack (already shipped): nx_tensor.nx -- NxTensor (Q10 i64 storage) nx_placement.nx -- NxPlacedTensor (Phase A: OWNED/MMAP/VRAM/NETWORK) This-agent stack (this session): nx_tensor_placement.nx -- 8-tier placement registry (refinement on Phase A) nx_dispatcher.nx -- Phase C op routing nx_resource_arbiter.nx -- Phase D budget tracking nx_actor.nx -- cooperative actor scheduler nx_message.nx -- typed inter-actor passing nx_stream_multiplexer.nx -- output stream interleaving nx_interrupt_broker.nx -- sub-50ms preemption Scenario: a 3-actor DND-companion mini-session. Actor 1 (LLM): generates tokens, emits LLM_TOKEN messages Actor 2 (TTS): subscribes to LLM_TOKEN, pushes AUDIO_FRAME chunks to mux Actor 3 (GAME): pushes GAME_ACTION chunks to mux Operator clicks -> interrupt broker yields LLM checkpoint within 50ms