@elizaos/plugin-aosp-local-inference

AOSP-only llama.cpp FFI bindings (via bun:ffi) and the local-inference bootstrap that registers TEXT_SMALL, TEXT_LARGE, TEXT_EMBEDDING, and TEXT_TO_SPEECH model handlers on the AOSP mobile agent.

Both exports self-gate on ELIZA_LOCAL_LLAMA=1 and are no-ops on every other platform/runtime, so they are safe to import unconditionally from the mobile agent's static plugin barrel.

Public surface

• registerAospLlamaLoader() — registers the bun:ffi-backed llama loader with @elizaos/agent when running on AOSP.
• ensureAospLocalInferenceHandlers() — registers the text / embedding handlers against the AOSP llama loader, plus a Kokoro-backed TEXT_TO_SPEECH handler that reuses @elizaos/shared/local-inference discovery/runtime code and emits WAV bytes.

Layout

plugins/plugin-aosp-local-inference/
  src/
    index.ts                              Barrel
    aosp-llama-adapter.ts                 bun:ffi loader registration
    aosp-llama-streaming.ts               Streaming bridge
    aosp-dflash-adapter.ts                DFlash drafter wiring
    aosp-local-inference-bootstrap.ts     Model-handler registrar incl. Kokoro TTS
  __tests__/                              vitest suites