API

ExecuTorch has two runtime layers:

• The core runtime loads and executes methods in a compatible .pte program. Use it when your application owns preprocessing, postprocessing, and task orchestration.

• The LLM runner builds on the core runtime with tokenization, prefill and decode orchestration, sampling, and streaming generation for text and multimodal models.

The table below is the entry point for choosing a language binding. Maturity follows the API Life Cycle and Deprecation Policy: an API is stable unless it is explicitly marked experimental or deprecated. An annotation on an individual API takes precedence over this summary.

Runtime and LLM APIs by language

Language / platform	Core runtime	LLM runner	Maturity
C++	Running an ExecuTorch Model Using the Module Extension in C++ for the high-level Module API, or Runtime API Reference for Program and Method	TextLLMRunner and MultimodalRunner	Core: stable; LLM: experimental
Python	Runtime Python API Reference for host-side loading, execution, and validation	Python LLM runner bindings, whose availability depends on the installed package or source build	Core: stable; LLM: experimental
Android: Java / Kotlin	Using ExecuTorch on Android and the Javadoc for Module, Tensor, and EValue	Running LLMs on Android for LlmModule	Core: experimental; LLM: experimental
Apple: Swift / Objective-C	Using ExecuTorch on iOS for Module, Tensor, and Value	Running LLMs on iOS for TextRunner and MultimodalRunner	Core: follows API Life Cycle and Deprecation Policy; LLM: experimental
Browser: JavaScript / WebAssembly	WebAssembly Module and Tensor, currently built from source	No high-level LLM runner API	Core: experimental

Use the core runtime for any compatible exported model, including vision, audio, and custom workloads. Use an LLM runner only when the exported program and tokenizer satisfy that runner’s model metadata and packaging requirements.

Export and reference documentation

• Export API Reference — Export to ExecuTorch API Reference

• Runtime API Reference — ExecuTorch Runtime API Reference

• Runtime Python API Reference — Runtime Python API Reference

• API Life Cycle and Deprecation Policy — API Life Cycle

• Android API reference: Java/Kotlin API documentation

• Running an ExecuTorch Model Using the Module Extension in C++ — Extension Module

• Managing Tensor Memory in C++ — Extension Tensor

• Detailed C++ Runtime APIs Tutorial — Detailed C++ Runtime APIs Tutorial